diff --git a/extensions/iterator.js b/extensions/iterator.js
index df9c841..7577a49 100644
--- a/extensions/iterator.js
+++ b/extensions/iterator.js
@@ -60,6 +60,8 @@ import {
 	completeFeatureWaveAbort,
 	extractPathsFromBash,
 	footerText,
+	implementationCommand,
+	implementationHandoffState,
 	mergePayload,
 	nextAutoAction,
 	nextFeatureWaveAction,
@@ -68,6 +70,7 @@ import {
 	roleFromInput,
 	roleModelSpec,
 	runJson,
+	shouldApplyRole,
 	scriptPath,
 	shouldNudge,
 	uiPort,
@@ -318,7 +321,10 @@ export default function iteratorExtension(pi) {
 	};
 
 	/** Refresh the idle dashboard tabs (no pending round). */
-	const refreshHub = async (cwd, { preferPlanning = false } = {}) => {
+	const refreshHub = async (
+		cwd,
+		{ preferPlanning = false, activateWork = false } = {},
+	) => {
 		if (!session || !session.isRunning() || session.hasPending()) return;
 		try {
 			// Inactive tabs first: their refreshes are stored silently. On first
@@ -340,7 +346,11 @@ export default function iteratorExtension(pi) {
 				render: () => VIEWS.planning({ ...hub, step: "planning" }),
 				activate: landOnPlanning,
 			});
-			session.showView({ step: "hub", render: () => VIEWS.hub(hub) });
+			session.showView({
+				step: "hub",
+				render: () => VIEWS.hub(hub),
+				activate: activateWork,
+			});
 			await pushStatus(cwd);
 		} catch {
 			/* a broken bundle read must never take pi down */
@@ -376,7 +386,8 @@ export default function iteratorExtension(pi) {
 					"info",
 				);
 			}
-			await refreshHub(cwd);
+			session?.closeModal?.();
+			await refreshStatus(cwd);
 		} catch (e) {
 			if (lastCtx?.hasUI)
 				lastCtx.ui.notify(
@@ -390,8 +401,8 @@ export default function iteratorExtension(pi) {
 	const saveBacklog = async (input) => {
 		const cwd = ctxCwd();
 		// Backlog writes are allowed during a model turn, but must not replace or
-		// clear that turn's working guard. The normal turn-end refresh will pick up
-		// the filesystem change; idle saves still refresh immediately.
+		// clear that turn's working guard. Refresh both stored dashboard tabs after
+		// every save; session.showView keeps the owned overlay intact.
 		const preserveAgentWorking = session?.isWorking?.() === true;
 		if (!preserveAgentWorking)
 			session?.showWorking?.("Saving backlog candidate…");
@@ -402,6 +413,7 @@ export default function iteratorExtension(pi) {
 					op: "backlog",
 					action: input.action,
 					id: input.id,
+					ids: input.ids,
 					title: input.title,
 					details: input.details,
 					kind: input.kind,
@@ -409,14 +421,65 @@ export default function iteratorExtension(pi) {
 				}),
 			});
 			invalidateSession();
-			notifyUi(`backlog ${result.action}d: ${result.item.title}`, "info");
+			const savedLabel =
+				result.action === "select-many"
+					? `${result.changedItems.length} backlog candidates updated`
+					: `backlog ${result.action}d: ${result.item.title}`;
+			notifyUi(savedLabel, "info");
 		} catch (e) {
 			notifyUi(`backlog not saved — ${e.message}`, "error");
 		} finally {
-			if (!preserveAgentWorking) {
-				session?.clearWorking?.();
+			if (!preserveAgentWorking) session?.clearWorking?.();
+			await refreshHub(cwd);
+		}
+	};
+
+	/** Review and apply a complete pasted plan without spending a planner turn. */
+	const startFastPlan = async (input) => {
+		const cwd = ctxCwd();
+		const draft = input?.structuredPlan;
+		if (
+			!draft?.title ||
+			!draft?.plan?.goal ||
+			!draft?.plan?.architecture ||
+			!draft?.plan?.keyDecisions
+		) {
+			notifyUi("structured plan is missing a required section", "error");
+			return;
+		}
+		try {
+			const gathered = await gatherPayload(cwd, "plan");
+			const payload = mergePayload(gathered, { ...draft, fastTrack: true });
+			const result = await session.showStep({
+				step: "plan",
+				render: () => VIEWS.plan(payload),
+			});
+			// This direct path has no planner agent waiting behind the review round.
+			session.clearWorking?.();
+			if (result?.type !== "plan-approved") {
 				await refreshHub(cwd);
+				return;
 			}
+			const applied = await runJson(scriptPath("write"), [], {
+				cwd,
+				stdin: JSON.stringify({
+					op: "plan",
+					title: payload.title,
+					...(payload.description ? { description: payload.description } : {}),
+					sections: result.sections,
+					dependencies: result.dependencies || [],
+				}),
+			});
+			invalidateSession();
+			if (!applied?.ok)
+				throw new Error(applied?.error || "plan was not applied");
+			await refreshHub(cwd, { activateWork: true });
+			session.showWorking("Plan approved — preparing feature breakdown…");
+			dispatch("/skill:iterator-feature");
+		} catch (e) {
+			session?.clearWorking?.();
+			notifyUi(`structured plan not saved — ${e.message}`, "error");
+			await refreshHub(cwd);
 		}
 	};
 
@@ -436,15 +499,18 @@ export default function iteratorExtension(pi) {
 		}
 	};
 
-	/** Show the settings view as an idle dashboard page (unsolicited round). */
+	/** Open Settings above the shell without replacing its tab or pending round. */
 	const openSettings = async () => {
 		const cwd = ctxCwd();
 		try {
 			const payload = await gatherPayload(cwd, "settings");
 			const models = await modelOptions();
-			session.showView({
-				step: "settings",
-				render: () => VIEWS.settings(models ? { ...payload, models } : payload),
+			session.showModal({
+				render: () =>
+					VIEWS.settings({
+						...(models ? { ...payload, models } : payload),
+						modal: true,
+					}),
 			});
 		} catch (e) {
 			if (lastCtx?.hasUI) lastCtx.ui.notify(`iterator: ${e.message}`, "error");
@@ -535,7 +601,9 @@ export default function iteratorExtension(pi) {
 	let featureWave = null; // fixed ready-feature snapshot; review stays manual
 	let preAutoModel = null; // the user's model before the first role switch
 	let pendingRole = null; // exact role command captured from the current input
-	let manualRoleActive = false; // this turn switched a manual role model
+	// A before_agent_start/agent_end pair can overlap a queued follow-up. Keep
+	// restoration ownership in FIFO order instead of a shared boolean.
+	const manualRoleTurns = [];
 
 	const notifyUi = (msg, level = "info") => {
 		if (lastCtx?.hasUI) lastCtx.ui.notify(`iterator: ${msg}`, level);
@@ -556,9 +624,43 @@ export default function iteratorExtension(pi) {
 			streamingBehavior: "followUp", // older runtimes take the raw option name
 		});
 
-	/** Apply a role's model/thinking overrides; remember the user's model once. */
+	const startConflictResolution = (result) => {
+		const feature = String(result?.feature || "").trim();
+		const target = String(result?.target || "").trim();
+		if (
+			!/^[a-z0-9][a-z0-9-]*$/.test(feature) ||
+			!/^[a-z-]+\/[a-z0-9][a-z0-9-]*$/.test(target)
+		) {
+			notifyUi("invalid feature or decision conflict target", "error");
+			return;
+		}
+		const note = String(result.prompt || "Recorded decision conflict")
+			.replace(/\s+/g, " ")
+			.trim()
+			.slice(0, 500);
+		const anchors = (Array.isArray(result.anchors) ? result.anchors : [])
+			.map(String)
+			.filter(Boolean)
+			.slice(0, 12)
+			.join(", ");
+		const followUp =
+			`Resolve the conflict recorded on feature ${feature}: ${note}. ` +
+			`Use the existing reviewed update-memory flow for ${target}; no durable knowledge may change before an explicit memory verdict. ` +
+			(anchors ? `Re-check against anchored files ${anchors}. ` : "") +
+			`After an accepted memory update, continue into /skill:iterator-feature with guidance to re-check only ${feature} against ${target}, preserving every unrelated feature and conflict flag.`;
+		session.showWorking("Opening reviewed decision conflict resolution…");
+		dispatch(`/skill:iterator-knowledge update-memory ${target} — ${followUp}`);
+	};
+
+	/**
+	 * Apply a role's model/thinking overrides and report whether the model
+	 * actually changed. Only a successful switch may arm restoration: a failed
+	 * provider lookup must leave the active session model (and its credentials)
+	 * completely untouched.
+	 */
 	const applyRole = async (role, settings) => {
 		const spec = roleModelSpec(settings, role);
+		let switchedModel = false;
 		try {
 			if (spec.model) {
 				const slash = spec.model.indexOf("/");
@@ -566,13 +668,17 @@ export default function iteratorExtension(pi) {
 				const id = spec.model.slice(slash + 1);
 				const m = lastCtx?.modelRegistry?.find?.(provider, id);
 				if (m) {
-					if (!preAutoModel) preAutoModel = lastCtx?.model || null;
+					const previousModel = preAutoModel || lastCtx?.model || null;
 					const ok = await pi.setModel(m);
-					if (!ok)
+					if (ok) {
+						if (!preAutoModel) preAutoModel = previousModel;
+						switchedModel = true;
+					} else {
 						notifyUi(
 							`no API key for ${spec.model} — staying on the active model`,
 							"warning",
 						);
+					}
 				} else {
 					notifyUi(
 						`unknown model ${spec.model} for ${role} — staying on the active model`,
@@ -587,6 +693,7 @@ export default function iteratorExtension(pi) {
 				"warning",
 			);
 		}
+		return switchedModel;
 	};
 
 	/** Restore the user's model after an automatic or manual role turn. */
@@ -650,7 +757,9 @@ export default function iteratorExtension(pi) {
 				phase: "implementing",
 				active_feature: action.feature,
 			});
-			await applyRole(action.role, settings);
+			// The replacement session applies the implementer role when its final
+			// skill command starts; switching here would leak a role model into a
+			// session that is about to be torn down.
 			attribution = { step: action.step, feature: action.feature };
 			session?.showWorking({
 				text: `Wave: implementing ${action.feature} (${featureWave.results.length}/${featureWave.results.length + featureWave.queue.length + 1} finished)…`,
@@ -658,7 +767,7 @@ export default function iteratorExtension(pi) {
 				feature: action.feature,
 			});
 			await pushStatus(cwd);
-			await dispatch(action.cmd);
+			await dispatch(implementationCommand(action.feature, { auto: true }));
 		} catch (error) {
 			featureWave = null;
 			await writeState({
@@ -768,29 +877,12 @@ export default function iteratorExtension(pi) {
 				});
 				invalidateSession();
 			}
-			// Fresh context per feature: when the next implement round targets a
-			// DIFFERENT feature than the previous one, clear the conversation
-			// context if the runtime supports it. Correctness never depends on
-			// this — the implement gather payload (contract, relevantMemories,
-			// finishedFeatures) is self-contained by design.
-			if (
-				action.step === "implement" &&
-				action.feature &&
-				sess.state?.active_feature &&
-				sess.state.active_feature !== action.feature
-			) {
-				try {
-					if (typeof pi.clearContext === "function") await pi.clearContext();
-					else if (typeof pi.compact === "function") await pi.compact();
-				} catch {
-					/* best-effort only */
-				}
-			}
 			await writeState({
 				phase: AUTO_PHASE_FOR_STEP[action.step] || "implementing",
 				active_feature: action.feature || null,
 			});
-			await applyRole(action.role, sess.settings);
+			if (action.step !== "implement")
+				await applyRole(action.role, sess.settings);
 			attribution = { step: action.step, feature: action.feature || null };
 			const p = sess.hub?.progress || {};
 			// Structured working state: the shell renders step/feature and a
@@ -803,7 +895,11 @@ export default function iteratorExtension(pi) {
 			});
 			await pushStatus(cwd);
 			try {
-				await dispatch(action.cmd);
+				await dispatch(
+					action.step === "implement"
+						? implementationCommand(action.feature, { auto: true })
+						: action.cmd,
+				);
 			} catch (err) {
 				// Recoverable, never a silent stall: keep mode:auto but pause, so
 				// the control strip's Continue re-enters kickAuto and re-dispatches.
@@ -932,7 +1028,10 @@ export default function iteratorExtension(pi) {
 			session?.showWorking?.("Resuming with your guidance…");
 			dispatch(
 				feature
-					? `/skill:iterator-implement ${feature} --auto — user guidance: ${guidance}`
+					? implementationCommand(feature, {
+							auto: true,
+							guidance: `user guidance: ${guidance}`,
+						})
 					: guidance,
 			);
 		} catch (e) {
@@ -972,8 +1071,12 @@ export default function iteratorExtension(pi) {
 						void saveUsagePrices(result.prices);
 						return;
 					}
-					// Settings is an idle page: its Close button emits cancel, which
-					// must restore the dashboard rather than leave its view in place.
+					// Settings is shell-owned: dismissal leaves the originating tab,
+					// pending round, and Work overlay untouched.
+					if (result?.type === "settings-close") {
+						session.closeModal();
+						return;
+					}
 					if (result?.type === "cancel") {
 						void refreshHub(ctxCwd());
 						return;
@@ -986,13 +1089,31 @@ export default function iteratorExtension(pi) {
 						void openArchive(result.feature);
 						return;
 					}
+					if (
+						result?.type === "action" &&
+						result.action === "plan-fast-track"
+					) {
+						void startFastPlan(result);
+						return;
+					}
+					if (
+						result?.type === "action" &&
+						result.action === "resolve-memory-conflict"
+					) {
+						startConflictResolution(result);
+						return;
+					}
 					// Knowledge's page-level Close mirrors Settings: back to Work
 					// (decisions/settings-close-returns-to-work) — never a model turn.
 					// "planning" is pure navigation too: refresh both dashboard tabs
 					// (the shell's Planning tab already holds the view).
+					if (result?.type === "action" && result.action === "hub") {
+						void refreshHub(ctxCwd(), { activateWork: true });
+						return;
+					}
 					if (
 						result?.type === "action" &&
-						["hub", "close", "planning"].includes(result.action)
+						["close", "planning"].includes(result.action)
 					) {
 						void refreshHub(ctxCwd());
 						return;
@@ -1051,24 +1172,62 @@ export default function iteratorExtension(pi) {
 	};
 
 	// ---------------------------------------------------------------------
-	// Friendly commands (skills stay the source of the flow logic).
-	// /iterator-implement with no argument turns into a TUI feature picker.
+	// Friendly commands (skills stay the source of the flow logic). An
+	// implementation is the one exception: it replaces the Pi session before
+	// sending the skill command, so the fresh agent sees the feature contract
+	// rather than the accumulated conversation.
+
+	const startImplementationSession = async (args, ctx) => {
+		const rawArgs = String(args || "").trim();
+		const [commandArgs, guidance] = rawArgs.split(/\s+—\s+/, 2);
+		const tokens = commandArgs.split(/\s+/).filter(Boolean);
+		let feature = tokens.find((token) => !token.startsWith("--")) || null;
+		const auto = tokens.includes("--auto");
+		if (!feature) {
+			const picked = await pickReadyFeature(ctx);
+			if (picked === undefined) return;
+			feature = picked;
+		}
+		const command = `/skill:iterator-implement ${feature}${auto ? " --auto" : ""}${guidance ? ` — ${guidance}` : ""}`;
+		const handoff = {
+			feature,
+			auto,
+			autoSteps,
+			// A ready-wave lives in extension memory; preserve only its plain,
+			// immutable snapshot so the replacement runtime can finish the wave.
+			featureWave: featureWave
+				? {
+						...featureWave,
+						queue: [...featureWave.queue],
+						results: [...featureWave.results],
+					}
+				: null,
+		};
+		// A tester/reviewer role belongs to the old session. Return to the user's
+		// model before replacement so an `active` implementer never inherits it.
+		await restoreModel();
+		const result = await ctx.newSession({
+			parentSession: ctx.sessionManager?.getSessionFile?.(),
+			setup: async (manager) => {
+				manager.appendCustomEntry("iterator-implementation-handoff", handoff);
+			},
+			withSession: async (replacementCtx) => {
+				await replacementCtx.sendUserMessage(command);
+			},
+		});
+		if (result?.cancelled && ctx.hasUI)
+			ctx.ui.notify("iterator: fresh implementation session cancelled", "info");
+	};
 
 	for (const command of COMMANDS) {
 		pi.registerCommand(command.name, {
 			description: command.description,
 			handler: async (args = "", ctx) => {
-				const trimmedArgs = String(args).trim();
-				if (
-					command.name === "iterator-implement" &&
-					!trimmedArgs &&
-					ctx?.hasUI
-				) {
-					const picked = await pickReadyFeature(ctx);
-					if (picked === undefined) return; // dismissed / nothing ready
-					dispatch(`/skill:iterator-implement ${picked}`.trim());
+				if (command.name === "iterator-implement") {
+					await startImplementationSession(args, ctx);
 					return;
 				}
+				const trimmedArgs = String(args).trim();
 				dispatch(
 					`/skill:${command.name}${trimmedArgs ? ` ${trimmedArgs}` : ""}`,
 				);
@@ -1092,7 +1251,7 @@ export default function iteratorExtension(pi) {
 
 	pi.registerCommand("iterator-next", {
 		description:
-			"Implement the next dependency-ready feature, no questions asked.",
+			"Implement the next dependency-ready feature in a fresh session.",
 		handler: async (_args, ctx) => {
 			try {
 				const imp = await gatherPayload(ctx.cwd, "implement");
@@ -1106,7 +1265,7 @@ export default function iteratorExtension(pi) {
 						ctx.ui.notify(`iterator: nothing to implement — ${why}`, "warning");
 					return;
 				}
-				dispatch(`/skill:iterator-implement ${imp.next.name}`);
+				await startImplementationSession(imp.next.name, ctx);
 			} catch (e) {
 				if (ctx.hasUI) ctx.ui.notify(`iterator: ${e.message}`, "error");
 			}
@@ -1247,7 +1406,13 @@ export default function iteratorExtension(pi) {
 				const approved =
 					(params.op === "adjustments" || params.type === "plan-approved") &&
 					(params.accept === true || params.type === "plan-approved");
+				if (params.op === "commit-tests" && result?.ok) {
+					// Red tests are an intentional implementation handoff: return to
+					// Work so their paths and target state are immediately visible.
+					void refreshHub(ctx.cwd, { activateWork: true });
+				}
 				if (approved) {
+					void refreshHub(ctx.cwd, { activateWork: true });
 					try {
 						const { settings, state } = await gatherSession(ctx.cwd);
 						if (settings?.auto_mode === "on" && state?.mode !== "auto") {
@@ -1465,6 +1630,7 @@ export default function iteratorExtension(pi) {
 						applied = { ok: false, error: e.message };
 					}
 					invalidateSession();
+					if (applied?.ok) await refreshHub(ctx.cwd, { activateWork: true });
 					return asText({ ...result, applied });
 				}
 				return asText(result);
@@ -1484,11 +1650,19 @@ export default function iteratorExtension(pi) {
 		rememberCtx(ctx);
 		const workOwner = session?.ensureWorking?.("AI is working…") ?? null;
 		agentWorkOwners.push(workOwner);
+		const manualRoleTurn = { switched: false };
+		manualRoleTurns.push(manualRoleTurn);
 		try {
 			const { hub, implement, settings, state } = await gatherSession(ctx.cwd);
-			if (pendingRole && state?.mode !== "auto" && !featureWave) {
-				await applyRole(pendingRole, settings);
-				manualRoleActive = true;
+			const role = pendingRole;
+			pendingRole = null; // role input belongs to exactly one agent turn
+			if (
+				shouldApplyRole(role, {
+					mode: state?.mode,
+					featureWave,
+				})
+			) {
+				manualRoleTurn.switched = await applyRole(role, settings);
 			}
 			// Model selection also applies to /iterator-plan before a bundle exists;
 			// only the ambient bundle context depends on durable plan state.
@@ -1571,8 +1745,14 @@ export default function iteratorExtension(pi) {
 	// plan-less project lands on Planning's goal/init hero rather than an empty
 	// Work tab.
 
-	pi.on("session_start", async (_event, ctx) => {
+	pi.on("session_start", async (event, ctx) => {
 		rememberCtx(ctx);
+		const sessionEntries = ctx.sessionManager?.getEntries?.() || [];
+		// A persisted marker is only a handoff during the session replacement
+		// that created it. On reload/restart, retain the normal auto safety pause.
+		const handoff = implementationHandoffState(sessionEntries, event?.reason);
+		if (handoff?.featureWave) featureWave = handoff.featureWave;
+		if (handoff) autoSteps = handoff.autoSteps;
 		await refreshStatus(ctx);
 		try {
 			await ensureServer(ctx);
@@ -1580,6 +1760,7 @@ export default function iteratorExtension(pi) {
 			// pause it and let the human press Continue in the dashboard.
 			const { state } = await gatherSession(ctx.cwd);
 			if (
+				!handoff &&
 				state?.mode === "auto" &&
 				!state.paused &&
 				["testing", "implementing", "reviewing"].includes(state.phase)
@@ -1609,13 +1790,11 @@ export default function iteratorExtension(pi) {
 	pi.on("agent_end", async (_event, ctx) => {
 		rememberCtx(ctx);
 		const endedWorkOwner = agentWorkOwners.shift() ?? null;
+		const manualRoleTurn = manualRoleTurns.shift();
 		try {
 			invalidateSession(); // the turn may have changed files/commits
 			await flushUsage(ctx.cwd);
-			if (manualRoleActive) {
-				manualRoleActive = false;
-				await restoreModel();
-			}
+			if (manualRoleTurn?.switched) await restoreModel();
 			await refreshHub(ctx.cwd);
 			await refreshStatus(ctx);
 			// Keep abortPending set until this stale agent_end reaches its final
diff --git a/lib/gather.mjs b/lib/gather.mjs
index 4394ada..f13f78e 100644
--- a/lib/gather.mjs
+++ b/lib/gather.mjs
@@ -395,6 +395,7 @@ export function gather(startDir) {
 	];
 
 	const trailerMap = featureCommitMap(b.root);
+	const conceptsById = new Map(loadConcepts(b.memDir).map((c) => [c.id, c]));
 	// Readiness is computed here, once, server-side — views only render it.
 	const ready = readiness(b.features, b.settings);
 	const features = b.features.map((c) => {
@@ -403,6 +404,17 @@ export function gather(startDir) {
 		const hasDiff = diffFiles.some((f) => globs.some((re) => re.test(f)));
 		const recorded = Array.isArray(c.fm.commits) && c.fm.commits.length > 0;
 		const hasCommits = recorded || trailerMap.has(c.slug);
+		const tests = listy(c.fm.tests);
+		const conflicts = featureConflicts(c.fm).map((conflict) => {
+			const decision = conceptsById.get(conflict.decision);
+			return {
+				...conflict,
+				title: decision?.title || conflict.decision,
+				description: decision?.description || "",
+				files: decision?.files || [],
+				decisionExists: Boolean(decision),
+			};
+		});
 		return {
 			name: c.slug,
 			title: c.fm.title || c.slug,
@@ -410,11 +422,13 @@ export function gather(startDir) {
 			status: c.fm.status || "pending",
 			size: c.fm.size || "small",
 			testsStatus: c.fm.tests_status || "none",
+			tests,
+			testCount: tests.length,
 			dependsOn: listy(c.fm.depends_on),
 			...ready.get(c.slug),
 			hasDiff,
 			hasCommits,
-			conflicts: featureConflicts(c.fm).length,
+			conflicts,
 		};
 	});
 
diff --git a/lib/pi-tools.mjs b/lib/pi-tools.mjs
index 7aefe12..3279f78 100644
--- a/lib/pi-tools.mjs
+++ b/lib/pi-tools.mjs
@@ -23,12 +23,61 @@ export function mergePayload(gathered, extra) {
  * Map a hub- or knowledge-view result to the command that runs the chosen
  * flow, or null for cancel/timeout/close/anything non-actionable.
  */
+export function implementationCommand(
+	feature,
+	{ auto = false, guidance = null } = {},
+) {
+	const parts = ["/iterator-implement"];
+	if (feature) parts.push(feature);
+	if (auto) parts.push("--auto");
+	if (guidance) parts.push(`— ${guidance}`);
+	return parts.join(" ");
+}
+
+/** Read and normalize the marker created for an intentional fresh session. */
+export function implementationHandoffState(entries, reason) {
+	if (reason !== "new" || !Array.isArray(entries)) return null;
+	for (let index = entries.length - 1; index >= 0; index -= 1) {
+		const entry = entries[index];
+		if (
+			entry?.type !== "custom" ||
+			entry.customType !== "iterator-implementation-handoff" ||
+			!entry.data ||
+			typeof entry.data !== "object"
+		)
+			continue;
+		return {
+			...entry.data,
+			autoSteps:
+				Number.isInteger(entry.data.autoSteps) && entry.data.autoSteps >= 0
+					? entry.data.autoSteps
+					: 0,
+		};
+	}
+	return null;
+}
+
+export function shouldApplyRole(role, { mode, featureWave } = {}) {
+	if (!role) return false;
+	// Auto/wave drivers no longer pre-apply the implementer role because the
+	// implementation runs after a session replacement. The resulting skill
+	// turn — including a harness-started direct skill invocation — owns it.
+	if (role === "implementer") return true;
+	return mode !== "auto" && !featureWave;
+}
+
 export function actionToCommand(result) {
 	if (!result || result.type !== "action") return null;
 
-	// Hub (Work tab): step flows, optionally scoped to a feature. A typed plan
-	// goal from the hero rides along so /iterator-plan can skip its questions.
-	const STEPS = ["plan", "feature", "test", "implement", "review", "design"];
+	// Implementation uses the extension command rather than a direct skill
+	// expansion: only a command context can create the focused replacement
+	// session before the skill gets its deterministic feature contract.
+	if (result.action === "implement")
+		return implementationCommand(result.feature, { guidance: result.prompt });
+
+	// Hub (Work tab): other step flows, optionally scoped to a feature. A typed
+	// plan goal from the hero rides along so /iterator-plan can skip questions.
+	const STEPS = ["plan", "feature", "test", "review", "design"];
 	if (STEPS.includes(result.action)) {
 		const parts = [`/skill:iterator-${result.action}`];
 		if (result.feature) parts.push(result.feature);
@@ -172,12 +221,23 @@ export function composeAmbientContext(hub, implement, concepts = []) {
 		const p = hub.progress || {};
 		const red = (hub.features || [])
 			.filter((c) => c.testsStatus === "red")
-			.map((c) => c.name);
+			.map((c) => ({
+				name: c.name,
+				tests: Array.isArray(c.tests) ? c.tests.slice(0, 3) : [],
+			}));
 		const parts = [
 			`Plan "${hub.plan.title}" — ${p.done ?? 0}/${p.total ?? 0} features done`,
 			`next ready: ${implement?.next?.name || "none"}`,
 		];
-		if (red.length) parts.push(`tests red: ${red.join(", ")}`);
+		if (red.length) {
+			parts.push(
+				`committed red tests: ${red
+					.map((c) =>
+						c.tests.length ? `${c.name} (${c.tests.join(", ")})` : c.name,
+					)
+					.join("; ")}`,
+			);
+		}
 		lines.push(
 			`iterator: ${parts.join(" · ")}. Route new implementation work through the feature flow (/iterator-implement or the iterator tools), not ad-hoc edits.`,
 		);
diff --git a/lib/session-server.mjs b/lib/session-server.mjs
index 3c5d634..38c83b0 100644
--- a/lib/session-server.mjs
+++ b/lib/session-server.mjs
@@ -144,6 +144,18 @@ iframe{display:block;width:100%;flex:1;border:0}
 #overlay-btns button#ov-abort:hover{border-color:#f85149}
 #conn{position:fixed;right:10px;bottom:10px;display:none;font:12px -apple-system,sans-serif;
   color:#d29922;background:#161b22;border:1px solid #30363d;border-radius:12px;padding:3px 10px;z-index:60}
+#settings-modal{position:fixed;inset:0;display:none;align-items:center;justify-content:center;
+  padding:var(--modal-gap,24px);background:rgba(1,4,9,.72);z-index:70}
+#settings-modal.open{display:flex}
+#settings-dialog{width:min(860px,100%);height:min(780px,100%);display:flex;flex-direction:column;
+  overflow:hidden;background:#0d1117;border:1px solid #30363d;border-radius:10px;box-shadow:0 12px 40px rgba(0,0,0,.5)}
+#settings-dialog header{display:flex;align-items:center;justify-content:space-between;gap:12px;padding:8px 12px;
+  color:#c9d1d9;background:#161b22;border-bottom:1px solid #30363d;font:600 13px -apple-system,BlinkMacSystemFont,"Segoe UI",sans-serif}
+#settings-dialog button{font:12px -apple-system,BlinkMacSystemFont,"Segoe UI",sans-serif;color:#c9d1d9;
+  background:#21262d;border:1px solid #30363d;border-radius:6px;padding:5px 10px;cursor:pointer}
+#settings-dialog button:hover{border-color:#8b949e}
+#settings-frame{display:block;width:100%;flex:1;border:0}
+@media(max-width:640px){#settings-modal{--modal-gap:0}#settings-dialog{width:100%;height:100%;border-radius:0;border-width:0}}
 </style>
 </head>
 <body>
@@ -176,12 +188,22 @@ iframe{display:block;width:100%;flex:1;border:0}
   </div>
 </div>
 </main>
+<div id="settings-modal" role="dialog" aria-modal="true" aria-label="Project settings" aria-hidden="true">
+  <section id="settings-dialog">
+    <header><span>Project settings</span><button id="settings-close" type="button">Close</button></header>
+    <iframe id="settings-frame" title="Project settings"></iframe>
+  </section>
+</div>
 <div id="conn">reconnecting…</div>
 <script>
 const frame = document.getElementById('v');
 const overlay = document.getElementById('overlay');
 const overlayText = document.getElementById('overlay-text');
 const conn = document.getElementById('conn');
+const settingsModal = document.getElementById('settings-modal');
+const settingsFrame = document.getElementById('settings-frame');
+const settingsClose = document.getElementById('settings-close');
+let modalFocus = null;
 let tab = ${JSON.stringify(activeTab)};
 let seq = 0;
 let working = null; // structured working state { text, step?, feature?, progress?, activity? } | null
@@ -216,6 +238,25 @@ function applyWorking() {
   document.getElementById('ov-pause').textContent = paused ? 'Continue' : 'Pause';
   overlay.style.display = 'flex';
 }
+function setModal(open, v) {
+  if(open){
+    modalFocus = document.activeElement;
+    settingsModal.classList.add('open');
+    settingsModal.setAttribute('aria-hidden', 'false');
+    settingsFrame.src = '/settings?v=' + (v != null ? v : Date.now());
+    settingsClose.focus();
+    return;
+  }
+  settingsModal.classList.remove('open');
+  settingsModal.setAttribute('aria-hidden', 'true');
+  settingsFrame.removeAttribute('src');
+  if(modalFocus && typeof modalFocus.focus === 'function') modalFocus.focus();
+  modalFocus = null;
+}
+function closeSettings(){
+  fetch('/submit', { method: 'POST', headers: { 'Content-Type': 'application/json' },
+    body: JSON.stringify({ type: 'settings-close' }) }).catch(() => {});
+}
 function setTab(t, v) {
   tab = t;
   document.getElementById('context').textContent = './' + tab;
@@ -248,6 +289,11 @@ es.addEventListener('working', e => {
   working = (d && typeof d === 'object') ? d : { text: String(d || 'AI is working…') };
   applyWorking();
 });
+es.addEventListener('modal', e => {
+  conn.style.display = 'none';
+  const d = JSON.parse(e.data || '{}');
+  setModal(!!d.open, d.v);
+});
 // Control strip: plan/branch/mode chips + pause/continue, fed by 'status'
 // events; buttons post deterministic /control actions handled by the
 // extension without a model turn.
@@ -288,6 +334,8 @@ function control(action) {
 }
 document.getElementById('ctl-pause').addEventListener('click', () => control(paused ? 'continue' : 'pause'));
 document.getElementById('ctl-gear').addEventListener('click', () => control('open-settings'));
+settingsClose.addEventListener('click', closeSettings);
+window.addEventListener('keydown', e => { if(e.key === 'Escape' && settingsModal.classList.contains('open')) closeSettings(); });
 document.getElementById('ov-pause').addEventListener('click', () => control(paused ? 'continue' : 'pause'));
 document.getElementById('ov-abort').addEventListener('click', () => control('abort'));
 es.onopen = () => { conn.style.display = 'none'; };
@@ -339,6 +387,8 @@ export function createSessionServer({ onUnsolicited, onControl, log } = {}) {
 	let pending = null; // { resolve, timer, onAbort, signal } | null
 	let cancelTimer = null;
 	let status = null; // last control-strip status, replayed on connect
+	let modalHtml = null; // shell-owned Settings document, independent of tabs
+	let modalSeq = 0;
 
 	const broadcast = (event, data) => {
 		const msg = `event: ${event}\ndata: ${JSON.stringify(data)}\n\n`;
@@ -439,6 +489,9 @@ export function createSessionServer({ onUnsolicited, onControl, log } = {}) {
 			res.write(`event: ${event}\ndata: ${JSON.stringify(data)}\n\n`);
 			if (status)
 				res.write(`event: status\ndata: ${JSON.stringify(status)}\n\n`);
+			res.write(
+				`event: modal\ndata: ${JSON.stringify({ open: modalHtml !== null, v: modalSeq })}\n\n`,
+			);
 			const heartbeat = setInterval(() => {
 				try {
 					res.write(": ping\n\n");
@@ -450,6 +503,9 @@ export function createSessionServer({ onUnsolicited, onControl, log } = {}) {
 				clearInterval(heartbeat);
 				sseClients.delete(res);
 			});
+		} else if (req.method === "GET" && url.pathname === "/settings") {
+			res.writeHead(200, { "Content-Type": "text/html; charset=utf-8" });
+			res.end(modalHtml || PLACEHOLDER);
 		} else if (req.method === "POST" && url.pathname === "/submit") {
 			// The run-id guard protects the pending ROUND from stale tabs; with no
 			// round pending, every visible view (including the inactive tab's
@@ -476,7 +532,8 @@ export function createSessionServer({ onUnsolicited, onControl, log } = {}) {
 				// Every other unsolicited action stays blocked to prevent concurrent agent
 				// flows from a stale or JS-disabled dashboard.
 				const backlogWrite = parsed.type === "backlog";
-				if (!pending && working != null && !backlogWrite) {
+				const modalClose = parsed.type === "settings-close";
+				if (!pending && working != null && !backlogWrite && !modalClose) {
 					say("rejected unsolicited /submit while working");
 					res.writeHead(409, { "Content-Type": "application/json" });
 					res.end('{"busy":true}');
@@ -484,7 +541,10 @@ export function createSessionServer({ onUnsolicited, onControl, log } = {}) {
 				}
 				res.writeHead(200, { "Content-Type": "application/json" });
 				res.end('{"ok":true}');
-				if (pending) {
+				const modalResult =
+					modalHtml !== null &&
+					(parsed.type === "settings" || parsed.type === "settings-close");
+				if (pending && !modalResult) {
 					// The interactive view temporarily hid this agent's overlay. Resume
 					// the same ownership claim so its eventual agent_end can release it.
 					setWorking("Sent to Claude — working…", pending.workingOwner);
@@ -686,6 +746,22 @@ export function createSessionServer({ onUnsolicited, onControl, log } = {}) {
 			return setWorking(textOrObj);
 		},
 
+		/** Open shell-owned Settings without replacing the active tab or round. */
+		showModal({ render }) {
+			modalHtml = render();
+			modalSeq += 1;
+			broadcast("modal", { open: true, v: modalSeq });
+		},
+
+		/** Dismiss Settings and reveal the untouched tab and pending round. */
+		closeModal() {
+			if (modalHtml === null) return false;
+			modalHtml = null;
+			modalSeq += 1;
+			broadcast("modal", { open: false, v: modalSeq });
+			return true;
+		},
+
 		/** Return the current claim, creating a generic one for manual turns. */
 		ensureWorking(textOrObj = "AI is working…") {
 			return working === null ? setWorking(textOrObj) : workingOwner;
diff --git a/lib/ui.mjs b/lib/ui.mjs
index c4eac2a..0cc346b 100644
--- a/lib/ui.mjs
+++ b/lib/ui.mjs
@@ -217,6 +217,10 @@ async function post(payload, okMsg, options){
 function mdToHtml(src){
   if(!src || !src.trim()) return '<span class="empty">Empty.</span>';
   const escc = s => s.replace(/&/g,'&amp;').replace(/</g,'&lt;').replace(/>/g,'&gt;');
+  // inline() already escaped &, <, and > before link matching; only quotes
+  // remain unsafe in the captured href. Escaping the whole URL again would
+  // turn &amp; into &amp;amp; and corrupt query parameters.
+  const attr = s => s.replace(/"/g,'&quot;').replace(/'/g,'&#39;');
   const inline = s => escc(s)
     .replace(/\`([^\`]+)\`/g, (m,c)=>'<code>'+c+'</code>')
     .replace(/\\*\\*([^*]+)\\*\\*/g, '<strong>$1</strong>')
@@ -224,7 +228,7 @@ function mdToHtml(src){
     .replace(/\\[([^\\]]+)\\]\\(([^)]+)\\)/g, function(m,t,u){
       // Only linkify http(s)/mailto/relative targets — never javascript: etc.
       return /^(https?:|mailto:|[\\/.#])/i.test(u)
-        ? '<a href="'+u+'" target="_blank" rel="noopener">'+t+'</a>' : m;
+        ? '<a href="'+attr(u)+'" target="_blank" rel="noopener">'+t+'</a>' : m;
     });
   const lines = src.split('\\n');
   let html='', i=0, inList=false, listTag='';
@@ -255,6 +259,50 @@ function mdToHtml(src){
   closeList();
   return html;
 }
+
+// Split markdown into heading-owned blocks, preserving fenced code verbatim.
+// Used by memory review to focus changed sections without trusting raw HTML.
+function markdownBlocks(src){
+  var lines = String(src == null ? '' : src).replace(/\\r/g, '').split('\\n');
+  var blocks = [], current = [], key = '$intro', labels = {}, fenced = false;
+  function push(){
+    var text = current.join('\\n').trim();
+    if(text) blocks.push({ key:key, text:text });
+    current = [];
+  }
+  lines.forEach(function(line){
+    var trim = line.trim();
+    if(trim.startsWith('\`\`\`') || trim.startsWith('~~~')) fenced = !fenced;
+    var match = fenced ? null : line.match(/^(#{1,6})\\s+(.+)$/);
+    if(match){
+      push();
+      var base = match[2].trim().toLowerCase();
+      labels[base] = (labels[base] || 0) + 1;
+      key = base + '#' + labels[base];
+    }
+    current.push(line);
+  });
+  push();
+  return blocks;
+}
+function markdownChangeHtml(next, previous, isCreate){
+  var before = markdownBlocks(previous);
+  var after = markdownBlocks(next);
+  var old = new Map(before.map((block)=> [block.key, block]));
+  var seen = new Set();
+  var html = after.map((block)=> {
+    var prior = old.get(block.key);
+    seen.add(block.key);
+    var kind = isCreate || !prior ? 'added' : (prior.text === block.text ? 'unchanged' : 'modified');
+    var label = kind === 'added' ? 'New section' : (kind === 'modified' ? 'Changed section' : 'Unchanged');
+    return '<section class="change-block '+kind+'"><div class="change-label">'+label+'</div>'+mdToHtml(block.text)+'</section>';
+  }).join('');
+  before.forEach((block)=> {
+    if(seen.has(block.key)) return;
+    html += '<section class="change-block removed"><div class="change-label">Removed section</div>'+mdToHtml(block.text)+'</section>';
+  });
+  return html || mdToHtml(next);
+}
 `;
 
 /* ------------------------------------------------------------------ *
diff --git a/lib/views/hub.mjs b/lib/views/hub.mjs
index 90dad21..7d45781 100644
--- a/lib/views/hub.mjs
+++ b/lib/views/hub.mjs
@@ -2,9 +2,10 @@
  * iterator: Work dashboard UI on the shared shell (../ui.mjs,
  * ../server.mjs). The execution surface of the flow: the active plan's
  * progress, dependency graph, and feature cards with their Test / Implement /
- * Review actions. Planning owns backlog and plan-lifecycle controls; active
- * plan context and feature management live here. Both surfaces render from
- * the same gather payload so they can never disagree about state.
+ * Review actions. Work owns active-plan progress, lifecycle controls, and
+ * feature management; Planning owns backlog, future planning, and archives.
+ * Both surfaces render from the same gather payload so they cannot disagree
+ * about state.
  *
  *   input:  { step:"hub", branch,
  *             plan: { title, status } | null,      // null = no bundle yet
@@ -14,13 +15,15 @@
  *             dirty: { count, files },             // working-tree files outside the bundle
  *             features: [ { name, title, description, status, size,
  *                         testsStatus,                  // none | red | green
+ *                         tests, testCount,              // recorded red/green test targets
  *                         dependsOn, ready, waitingOn,  // server-computed readiness
  *                         hasDiff, hasCommits,
- *                         conflicts } ] }               // # of flagged decision conflicts
+ *                         conflicts } ] }               // flagged decisions + notes/anchors
  *             retired: [ { name, title, created } ]   // archived plans, newest first
  *   output: one JSON line to stdout —
- *     { type:"action", action:"planning"|"test"|"implement"|"review"|"review-all"|"implement-wave"|"auto-implement"
- *       |"cancel-feature"|"escalation-restart"|"escalation-guide",
+ *     { type:"action", action:"planning"|"plan"|"feature"|"test"|"implement"|"review"|"review-all"
+ *       |"review-plan"|"retire"|"implement-wave"|"auto-implement"|"cancel-plan"|"cancel-feature"
+ * *       |"escalation-restart"|"escalation-guide"|"resolve-memory-conflict",
  *       feature:"<slug>"|null,
  *       prompt:"<guidance>"|null }                 // escalation-guide only
  *     plus the shared { type:"cancel" } / { type:"timeout" }.
@@ -35,6 +38,8 @@ const CSS = `
 .escalation .er{font-size:var(--fs-sm);white-space:pre-wrap}
 .escalation .em{font-size:var(--fs-xs);color:var(--text-muted)}
 .escalation textarea{width:100%;min-height:56px;resize:vertical;background:var(--bg);border:1px solid var(--border);border-radius:var(--radius-sm);padding:8px 10px;color:var(--text);font:inherit;font-size:var(--fs-sm)}
+.conflict-list{display:grid;gap:var(--sp-2);margin:8px 0 10px}.conflict-item{display:flex;align-items:flex-start;justify-content:space-between;gap:var(--sp-3);padding:8px 10px;background:var(--bg-red);border-left:3px solid var(--dot-red);border-radius:0 var(--radius-sm) var(--radius-sm) 0}.conflict-copy{font-size:var(--fs-xs);line-height:1.45}.conflict-copy strong{display:block;color:var(--del-fg)}.conflict-copy code{font-size:11px}.conflict-action{flex-shrink:0}
+@media(max-width:640px){.conflict-item{flex-direction:column}.conflict-action{width:100%}}
 `;
 
 const BODY = `
@@ -99,8 +104,8 @@ function render(){
     w.appendChild(box);
   }
 
-  // Plan bar: progress + the execution controls. Lifecycle buttons (revise,
-  // feature, review-plan, retire, cancel) live on the Planning surface.
+  // Work keeps active-plan progress, lifecycle, and execution controls in one
+  // plan bar; Planning remains a staging and archive surface.
   const done = (D.progress&&D.progress.done)||0, total = (D.progress&&D.progress.total)||CH.length;
   const bar = document.createElement('div');
   bar.className = 'planbar';
@@ -110,6 +115,52 @@ function render(){
       ? '<span class="pcount">'+done+' / '+total+' features done</span>'+
         '<div class="pbar"><div style="width:'+Math.round(done/total*100)+'%"></div></div>'
       : '<span class="pcount">not broken into features yet</span>');
+  // Active-plan lifecycle belongs with its progress and feature work. Every
+  // condition is the server-derived stage; this view never infers it.
+  const revise = document.createElement('button');
+  revise.className = 'act';
+  revise.textContent = 'Revise plan';
+  revise.addEventListener('click', () => action('plan', null, 'Starting /iterator-plan'));
+  bar.appendChild(revise);
+  const refeature = document.createElement('button');
+  refeature.className = D.stage==='needs-features' ? 'act primary-act' : 'act';
+  refeature.textContent = D.stage==='needs-features' ? 'Feature the plan' : 'Re-feature';
+  refeature.title = D.stage==='needs-features'
+    ? 'Next step: break the approved plan into small, dependency-ordered features (/iterator-feature)'
+    : 'Redraw the feature set from the plan (/iterator-feature)';
+  refeature.addEventListener('click', () => action('feature', null, 'Starting /iterator-feature'));
+  bar.appendChild(refeature);
+  if(D.stage==='awaiting-plan-review' || D.stage==='retirable'){
+    const reviewPlan = document.createElement('button');
+    reviewPlan.className = 'act primary-act';
+    reviewPlan.textContent = D.plan.planReviewed ? 'Re-review plan' : 'Review plan';
+    reviewPlan.title = D.plan.planReviewed
+      ? 'Plan reviewed '+D.plan.planReviewed+' \\u2014 run the whole-plan review again'
+      : 'Review all changes and commits against the plan\\u2019s goals and decisions';
+    reviewPlan.addEventListener('click', () => action('review-plan', null, 'Starting /iterator-review-plan'));
+    bar.appendChild(reviewPlan);
+  }
+  if(D.stage==='retirable'){
+    const retire = document.createElement('button');
+    retire.className = 'act primary-act';
+    retire.textContent = 'Retire plan';
+    retire.title = 'Condense the finished plan into a decisions/ memory and archive its features';
+    confirmButton(retire, 'Retires the plan \\u2014 click again', () =>
+      action('retire', null, 'Starting plan retirement'));
+    bar.appendChild(retire);
+  }
+  const cancelPlan = document.createElement('button');
+  cancelPlan.className = 'act danger';
+  cancelPlan.textContent = 'Cancel plan';
+  const dirtyWarn = (D.dirty && D.dirty.count)
+    ? D.dirty.count + ' uncommitted file' + (D.dirty.count!==1?'s':'') + ' + '
+    : '';
+  cancelPlan.title = 'Abandon this plan: archive it and DELETE its branch/worktree'
+    + (dirtyWarn ? ' \\u2014 \\u26a0 ' + dirtyWarn + 'unmerged commits will be lost' : '');
+  confirmButton(cancelPlan, '\\u26a0 Deletes ' + dirtyWarn + 'branch \\u2014 click again', () =>
+    action('cancel-plan', null, 'Cancelling plan'));
+  bar.appendChild(cancelPlan);
+
   // Auto mode: once the feature set is approved (pending features exist), the
   // whole test → implement → review loop can run agent-driven.
   const readyWave = Array.isArray(D.readyWave) ? D.readyWave : [];
@@ -176,7 +227,13 @@ function makeCard(c){
   const ready = c.ready !== false;
   const draft = c.status==='draft';
   const implemented = c.status==='implemented';
+  const conflictDetails = Array.isArray(c.conflicts) ? c.conflicts : [];
+  const conflictCount = conflictDetails.length || Number(c.conflicts) || 0;
   const icon = '<i class="st '+(c.status==='done'?'done':draft?'draft':implemented?'implemented':'pending')+'"></i>';
+  const redTarget = c.testsStatus==='red'
+    ? '<div class="deps"><strong>Committed red tests — implementation target:</strong> '+
+      ((c.tests&&c.tests.length) ? c.tests.map(t=>'<code>'+esc(t)+'</code>').join(' ') : esc(String(c.testCount||0))+' test'+(c.testCount===1?'':'s'))+'</div>'
+    : '';
   card.innerHTML =
     '<div class="ch"><span class="cn">'+icon+esc(c.title||c.name)+'</span>'+
       '<span class="chip cmut">'+esc(c.name)+'</span>'+
@@ -184,16 +241,38 @@ function makeCard(c){
       (implemented?'<span class="chip cy">implemented \\u2014 awaiting review</span>':'')+
       (implemented&&!c.hasDiff&&!c.hasCommits?'<span class="chip cr" title="Marked implemented, but review would find nothing: no working-tree diff and no recorded commits. Usual causes: the work was committed outside the accept flow (no Feature: trailer) or landed in a different checkout (plan worktree vs main). Check git log for its files, or Restart the feature.">\\u26a0 no recorded changes</span>':'')+
       sizeChip(c)+testBadge(c)+
-      (c.conflicts?'<span class="chip cr" title="This feature contradicts a project decision — check its Decision conflicts section">\\u26a0 '+c.conflicts+' decision conflict'+(c.conflicts!==1?'s':'')+'</span>':'')+'</div>'+
-    '<div class="cdesc">'+esc(c.description||'')+'</div>'+
+      (conflictCount?'<span class="chip cr" title="This feature contradicts a project decision — review it before implementation">\\u26a0 '+conflictCount+' decision conflict'+(conflictCount!==1?'s':'')+'</span>':'')+'</div>'+
+    '<div class="cdesc">'+esc(c.description||'')+'</div>'+redTarget+
     ((c.dependsOn&&c.dependsOn.length)?'<div class="deps">depends on '+c.dependsOn.map(d=>'<code>'+esc(d)+'</code>').join(' ')+'</div>':'');
 
+  if(conflictDetails.length){
+    const list = document.createElement('div'); list.className = 'conflict-list';
+    conflictDetails.forEach(conflict => {
+      const row = document.createElement('div'); row.className = 'conflict-item';
+      const copy = document.createElement('div'); copy.className = 'conflict-copy';
+      const title = document.createElement('strong'); title.textContent = conflict.title || conflict.decision;
+      const note = document.createElement('div'); note.textContent = conflict.note || conflict.description || 'Recorded decision conflict';
+      const id = document.createElement('code'); id.textContent = conflict.decision;
+      copy.append(title, note, id);
+      const resolve = document.createElement('button'); resolve.className = 'act conflict-action'; resolve.type = 'button'; resolve.textContent = 'Review decision';
+      resolve.title = 'Draft a decision update, require memory review, then re-check only this feature conflict';
+      resolve.addEventListener('click', () => post({
+        type:'action', action:'resolve-memory-conflict', feature:c.name,
+        target:conflict.decision, prompt:conflict.note || null,
+        anchors:Array.isArray(conflict.files) ? conflict.files : []
+      }, 'Starting reviewed decision update'));
+      row.append(copy, resolve); list.appendChild(row);
+    });
+    card.appendChild(list);
+  }
+
   const btns = document.createElement('div');
   btns.className = 'btns';
 
   const impl = document.createElement('button');
   impl.className = 'act primary-act';
-  impl.textContent = 'Implement';
+  impl.textContent = c.testsStatus==='red' ? 'Drive tests green' : 'Implement';
+  if(c.testsStatus==='red') impl.title = 'Run the committed, intentionally failing tests and make them pass';
   if(c.status==='done'){ impl.disabled = true; impl.title = 'Already done'; }
   else if(implemented){ impl.disabled = true; impl.title = 'Implemented — review it'; }
   else if(draft){ impl.disabled = true; impl.title = 'Draft — accept the feature set first (Re-feature on the Planning tab)'; }
@@ -205,11 +284,14 @@ function makeCard(c){
 
   const test = document.createElement('button');
   test.className = 'act';
-  test.textContent = c.status==='pending' ? 'Test (red)' : 'Test';
-  test.title = c.status==='pending'
-    ? 'Write failing tests from the feature contract before implementing'
-    : 'Write passing tests against the implemented feature';
-  if(draft){ test.disabled = true; test.title = 'Draft — accept the feature set first'; }
+  test.textContent = c.testsStatus==='red' ? 'Tests committed (red)' : (c.status==='pending' ? 'Test (red)' : 'Test');
+  test.title = c.testsStatus==='red'
+    ? 'Committed red tests are the implementation target — drive them green'
+    : c.status==='pending'
+      ? 'Write failing tests from the feature contract before implementing'
+      : 'Write passing tests against the implemented feature';
+  if(c.testsStatus==='red'){ test.disabled = true; }
+  else if(draft){ test.disabled = true; test.title = 'Draft — accept the feature set first'; }
   test.addEventListener('click', () => action('test', c.name, 'Starting /iterator-test'));
 
   const rev = document.createElement('button');
diff --git a/lib/views/memory-review.mjs b/lib/views/memory-review.mjs
index 8ca0d29..8a1642f 100644
--- a/lib/views/memory-review.mjs
+++ b/lib/views/memory-review.mjs
@@ -22,74 +22,98 @@
  * change nothing) | keep (leave the existing file untouched) | delete (remove
  * the existing concept file).
  */
-import { renderPage, escHtml } from '../ui.mjs';
+import { renderPage, escHtml } from "../ui.mjs";
 
 const ACTION_BADGE = {
-  create: ['new', 'NEW'],
-  update: ['upd', 'UPDATE'],
-  delete: ['del', 'DELETE'],
-  keep: ['keep', 'KEEP'],
+	create: ["new", "NEW"],
+	update: ["upd", "UPDATE"],
+	delete: ["del", "DELETE"],
+	keep: ["keep", "KEEP"],
 };
 
 function banner(data, memories) {
-  const mode = data.mode;
-  const staleCount = memories.filter(m => m.stale).length;
-  const conflictCount = memories.filter(m => m.conflict).length;
-  const draftCount = memories.filter(m => m.action !== 'keep').length;
-  if (mode === 'init') {
-    return `Drafted <strong>${memories.length}</strong> memories from codebase analysis — review, comment, accept.`;
-  }
-  if (mode === 'consolidate') {
-    return `Reviewing <strong>${memories.length}</strong> existing memories`
-      + (staleCount ? ` — <span class="amber"><strong>${staleCount}</strong> stale</span>` : ' — none stale')
-      + '.';
-  }
-  const commitCount = Number(data.commitCount) || 0;
-  const draftLabel = draftCount === 1 ? 'draft' : 'drafts';
-  const commitLabel = commitCount === 1 ? 'commit' : 'commits';
-  const conflictLabel = conflictCount === 1 ? 'conflict' : 'conflicts';
-  const conflictHtml = conflictCount
-    ? ` — <span class="red"><strong>${conflictCount}</strong> ${conflictLabel}</span>`
-    : ' — no conflicts';
-  return `<strong>${draftCount}</strong> ${draftLabel} from `
-    + `<strong>${commitCount || '?'}</strong> ${commitLabel}` + conflictHtml + '.';
+	const mode = data.mode;
+	const staleCount = memories.filter((m) => m.stale).length;
+	const conflictCount = memories.filter((m) => m.conflict).length;
+	const draftCount = memories.filter((m) => m.action !== "keep").length;
+	if (mode === "init") {
+		return `Drafted <strong>${memories.length}</strong> memories from codebase analysis — review, comment, accept.`;
+	}
+	if (mode === "consolidate") {
+		return (
+			`Reviewing <strong>${memories.length}</strong> existing memories` +
+			(staleCount
+				? ` — <span class="amber"><strong>${staleCount}</strong> stale</span>`
+				: " — none stale") +
+			"."
+		);
+	}
+	const commitCount = Number(data.commitCount) || 0;
+	const draftLabel = draftCount === 1 ? "draft" : "drafts";
+	const commitLabel = commitCount === 1 ? "commit" : "commits";
+	const conflictLabel = conflictCount === 1 ? "conflict" : "conflicts";
+	const conflictHtml = conflictCount
+		? ` — <span class="red"><strong>${conflictCount}</strong> ${conflictLabel}</span>`
+		: " — no conflicts";
+	return (
+		`<strong>${draftCount}</strong> ${draftLabel} from ` +
+		`<strong>${commitCount || "?"}</strong> ${commitLabel}` +
+		conflictHtml +
+		"."
+	);
 }
 
 function verdictButtons(m) {
-  const btn = (verdict, label) =>
-    `<button type="button" data-verdict="${verdict}">${label}</button>`;
-  if (m.action === 'keep') return btn('keep', 'Keep') + btn('delete', 'Delete');
-  if (m.action === 'delete') return btn('delete', 'Delete') + btn('keep', 'Keep');
-  return btn('accept', 'Accept') + btn('reject', 'Reject');
+	const btn = (verdict, label) =>
+		`<button type="button" data-verdict="${verdict}">${label}</button>`;
+	if (m.action === "keep") return btn("keep", "Keep") + btn("delete", "Delete");
+	if (m.action === "delete")
+		return btn("delete", "Delete") + btn("keep", "Keep");
+	return btn("accept", "Accept") + btn("reject", "Reject");
 }
 
 function cardHtml(m) {
-  const [cls, label] = ACTION_BADGE[m.action] || ACTION_BADGE.keep;
-  const tags = Array.isArray(m.tags) && m.tags.length
-    ? ' · ' + m.tags.map(t => `<span class="tag">${escHtml(t)}</span>`).join(' ')
-    : '';
-  const commits = Array.isArray(m.sourceCommits) && m.sourceCommits.length
-    ? ` · <span class="commits">${escHtml(m.sourceCommits.map(c => String(c).slice(0, 7)).join(', '))}</span>`
-    : '';
-  const showExisting = m.existingBody !== null && m.existingBody !== undefined
-    && ['update', 'delete', 'keep'].includes(m.action);
-  return `<article class="card" data-id="${escHtml(m.id)}">
+	const [cls, label] = ACTION_BADGE[m.action] || ACTION_BADGE.keep;
+	const tags =
+		Array.isArray(m.tags) && m.tags.length
+			? " · " +
+				m.tags.map((t) => `<span class="tag">${escHtml(t)}</span>`).join(" ")
+			: "";
+	const commits =
+		Array.isArray(m.sourceCommits) && m.sourceCommits.length
+			? ` · <span class="commits">${escHtml(m.sourceCommits.map((c) => String(c).slice(0, 7)).join(", "))}</span>`
+			: "";
+	const showExisting =
+		m.existingBody !== null &&
+		m.existingBody !== undefined &&
+		["update", "delete", "keep"].includes(m.action);
+	return `<article class="card" data-id="${escHtml(m.id)}">
   <div class="card-head">
     <span class="card-title">${escHtml(m.title || m.id)}</span>
-    <span class="chip">${escHtml(m.type || '')}</span>
+    <span class="chip">${escHtml(m.type || "")}</span>
     <span class="badge ${cls}">${label}</span>
-    ${m.stale ? '<span class="badge stale">STALE</span>' : ''}
-    ${m.conflict ? `<button type="button" class="badge conflict conflict-btn" data-with="${escHtml(m.conflict.with || '')}">CONFLICT ⇄</button>` : ''}
+    ${m.stale ? '<span class="badge stale">STALE</span>' : ""}
+    ${m.conflict ? `<button type="button" class="badge conflict conflict-btn" data-with="${escHtml(m.conflict.with || "")}">CONFLICT ⇄</button>` : ""}
   </div>
   <div class="meta"><code>${escHtml(m.id)}.md</code>${tags}${commits}</div>
-  ${m.description ? `<div class="desc">${escHtml(m.description)}</div>` : ''}
-  ${m.stale && Array.isArray(m.staleReasons) && m.staleReasons.length
-    ? `<div class="stale-panel">${m.staleReasons.map(r => `<div>⚠ ${escHtml(r)}</div>`).join('')}</div>` : ''}
-  ${m.conflict
-    ? `<div class="conflict-panel" hidden>Conflicts with <code>${escHtml(m.conflict.with || '')}</code> — ${escHtml(m.conflict.summary || '')}</div>` : ''}
+  ${m.description ? `<div class="desc">${escHtml(m.description)}</div>` : ""}
+  ${m.reason ? `<div class="request-context"><strong>Why this change:</strong> ${escHtml(m.reason)}</div>` : ""}
+  ${
+		m.stale && Array.isArray(m.staleReasons) && m.staleReasons.length
+			? `<div class="stale-panel">${m.staleReasons.map((r) => `<div>⚠ ${escHtml(r)}</div>`).join("")}</div>`
+			: ""
+	}
+  ${
+		m.conflict
+			? `<div class="conflict-panel" hidden>Conflicts with <code>${escHtml(m.conflict.with || "")}</code> — ${escHtml(m.conflict.summary || "")}</div>`
+			: ""
+	}
   <div class="body md" data-body-id="${escHtml(m.id)}"></div>
-  ${showExisting
-    ? `<details class="existing"><summary>Current version on disk</summary><div class="md" data-existing-id="${escHtml(m.id)}"></div></details>` : ''}
+  ${
+		showExisting
+			? `<details class="existing"><summary>Current version on disk</summary><div class="md" data-existing-id="${escHtml(m.id)}"></div></details>`
+			: ""
+	}
   <div class="controls">
     <div class="verdicts" data-id="${escHtml(m.id)}">${verdictButtons(m)}</div>
     <textarea class="comment" data-id="${escHtml(m.id)}" rows="1"
@@ -99,27 +123,27 @@ function cardHtml(m) {
 }
 
 function areaSections(data, memories) {
-  const areas = Array.isArray(data.areas) ? data.areas : [];
-  const byArea = new Map(areas.map(a => [a.id, []]));
-  const other = [];
-  for (const m of memories) {
-    (byArea.get(m.area) || other).push(m);
-  }
-  let html = '';
-  for (const a of areas) {
-    const cards = byArea.get(a.id);
-    if (!cards.length && data.mode !== 'init') continue;
-    html += `<section class="area">
+	const areas = Array.isArray(data.areas) ? data.areas : [];
+	const byArea = new Map(areas.map((a) => [a.id, []]));
+	const other = [];
+	for (const m of memories) {
+		(byArea.get(m.area) || other).push(m);
+	}
+	let html = "";
+	for (const a of areas) {
+		const cards = byArea.get(a.id);
+		if (!cards.length && data.mode !== "init") continue;
+		html += `<section class="area">
 <h2>${escHtml(a.title || a.id)} <span class="count">${cards.length}</span></h2>
-${a.description ? `<p class="area-desc">${escHtml(a.description)}</p>` : ''}
-${cards.length ? cards.map(cardHtml).join('\n') : '<p class="empty">No memories drafted for this area.</p>'}
+${a.description ? `<p class="area-desc">${escHtml(a.description)}</p>` : ""}
+${cards.length ? cards.map(cardHtml).join("\n") : '<p class="empty">No memories drafted for this area.</p>'}
 </section>`;
-  }
-  if (other.length) {
-    html += `<section class="area"><h2>Other <span class="count">${other.length}</span></h2>
-${other.map(cardHtml).join('\n')}</section>`;
-  }
-  return html;
+	}
+	if (other.length) {
+		html += `<section class="area"><h2>Other <span class="count">${other.length}</span></h2>
+${other.map(cardHtml).join("\n")}</section>`;
+	}
+	return html;
 }
 
 const CSS = `
@@ -149,9 +173,15 @@ const CSS = `
 .meta code{font-size:12px}
 .tag{border:1px solid var(--border);border-radius:8px;padding:0 6px;font-size:11px}
 .desc{font-size:13px;color:var(--text-muted);font-style:italic;margin:2px 0 6px}
+.request-context{font-size:var(--fs-xs);line-height:1.45;background:var(--accent-soft);border-left:3px solid var(--accent);padding:6px 9px;margin:6px 0}.request-context strong{color:var(--accent)}
 .stale-panel{border-left:3px solid var(--dot-yellow);padding:4px 10px;margin:6px 0;font-size:13px;color:var(--dot-yellow)}
 .conflict-panel{border-left:3px solid var(--dot-red);padding:6px 10px;margin:6px 0;font-size:13px;color:var(--dot-red)}
 .body.md{border-top:1px solid var(--border);margin-top:8px;padding-top:4px;font-size:14px;overflow:hidden}
+.change-block{position:relative;margin:8px 0;padding:8px 10px;border-left:3px solid transparent;border-radius:0 var(--radius-sm) var(--radius-sm) 0}
+.change-block.added,.change-block.modified{background:var(--add-bg);border-left-color:var(--dot-green)}
+.change-block.removed{background:var(--del-bg);border-left-color:var(--dot-red);color:var(--del-fg)}
+.change-block.unchanged{padding-block:2px;opacity:.72}.change-label{font:600 10px var(--font-mono);letter-spacing:.05em;text-transform:uppercase;color:var(--text-muted);margin-bottom:4px}
+.change-block.added .change-label,.change-block.modified .change-label{color:var(--add-fg)}.change-block.removed .change-label{color:var(--del-fg)}
 .body.md.collapsed{max-height:280px;-webkit-mask-image:linear-gradient(black 75%,transparent);mask-image:linear-gradient(black 75%,transparent)}
 .show-more{display:block;margin:4px 0;font-size:12px;padding:2px 10px}
 details.existing{margin:8px 0;font-size:14px}
@@ -171,6 +201,7 @@ footer.general{margin:30px 0 60px}
 footer.general label{display:block;font-weight:600;margin-bottom:6px}
 #general-comment{width:100%;min-height:70px}
 .hint{color:var(--text-muted);font-size:12px;margin-top:6px}
+@media(max-width:640px){.change-block{margin-inline:0;padding:7px 8px}.controls{flex-direction:column}.verdicts{max-width:100%}}
 `;
 
 // Step client JS. No backticks / dollar-brace; data only via D.
@@ -206,7 +237,10 @@ function onReady() {
     var el = document.querySelector('[data-body-id="' + CSS.escape(m.id) + '"]');
     if (el) {
       var main = (m.action === 'keep' || m.body == null) ? (m.existingBody != null ? m.existingBody : m.body) : m.body;
-      el.innerHTML = mdToHtml(main == null ? '' : main);
+      var focusChanges = (m.action === 'create' || m.action === 'update') && m.body != null;
+      el.innerHTML = focusChanges
+        ? markdownChangeHtml(m.body, m.existingBody == null ? '' : m.existingBody, m.action === 'create')
+        : mdToHtml(main == null ? '' : main);
     }
     var ex = document.querySelector('[data-existing-id="' + CSS.escape(m.id) + '"]');
     if (ex) ex.innerHTML = mdToHtml(m.existingBody == null ? '' : m.existingBody);
@@ -257,16 +291,20 @@ onReady();
 `;
 
 export function render(data) {
-  const mode = ['init', 'consolidate', 'memorize'].includes(data.mode) ? data.mode : 'init';
-  const memories = Array.isArray(data.memories) ? data.memories : [];
-  const round = data.round || 1;
+	const mode = ["init", "consolidate", "memorize"].includes(data.mode)
+		? data.mode
+		: "init";
+	const memories = Array.isArray(data.memories) ? data.memories : [];
+	const round = data.round || 1;
 
-  const subtitleParts = [`/ ${mode} review`, `round ${round}`];
-  if (mode === 'memorize' && data.baseCommit && data.headCommit) {
-    subtitleParts.push(`${String(data.baseCommit).slice(0, 7)}..${String(data.headCommit).slice(0, 7)}`);
-  }
+	const subtitleParts = [`/ ${mode} review`, `round ${round}`];
+	if (mode === "memorize" && data.baseCommit && data.headCommit) {
+		subtitleParts.push(
+			`${String(data.baseCommit).slice(0, 7)}..${String(data.headCommit).slice(0, 7)}`,
+		);
+	}
 
-  const BODY = `<div class="wrap">
+	const BODY = `<div class="wrap">
 <div class="banner">${banner({ ...data, mode }, memories)}</div>
 <div id="areas">
 ${areaSections({ ...data, mode }, memories)}
@@ -280,16 +318,20 @@ ${areaSections({ ...data, mode }, memories)}
 </footer>
 </div>`;
 
-  return renderPage({
-    step: 'memory-review',
-    subtitle: subtitleParts.join(' · '),
-    branch: data.branch,
-    title: `${mode} review`,
-    data: { mode, memories, areas: Array.isArray(data.areas) ? data.areas : [] },
-    css: CSS,
-    body: BODY,
-    clientJs: JS,
-    primaryIdle: 'Accept',
-    primaryChanged: 'Send review',
-  });
+	return renderPage({
+		step: "memory-review",
+		subtitle: subtitleParts.join(" · "),
+		branch: data.branch,
+		title: `${mode} review`,
+		data: {
+			mode,
+			memories,
+			areas: Array.isArray(data.areas) ? data.areas : [],
+		},
+		css: CSS,
+		body: BODY,
+		clientJs: JS,
+		primaryIdle: "Accept",
+		primaryChanged: "Send review",
+	});
 }
diff --git a/lib/views/plan.mjs b/lib/views/plan.mjs
index 6b6f684..470bb12 100644
--- a/lib/views/plan.mjs
+++ b/lib/views/plan.mjs
@@ -11,12 +11,14 @@
  *     { type:"plan-approved"|"plan-feedback", sections, dependencies, comments, comment }
  *     plus the shared { type:"cancel" } / { type:"timeout" }.
  */
-import { renderPage } from '../ui.mjs';
+import { renderPage } from "../ui.mjs";
 
 const PLAN_CSS = `
 .main{max-width:960px;margin:0 auto;padding:28px var(--sp-5)}
 h1{font-family:var(--font-display);font-size:var(--fs-xl);font-weight:600;margin-bottom:6px}
 .hint{font-size:var(--fs-sm);color:var(--text-muted);margin-bottom:28px;line-height:1.5}
+.fast-track{display:inline-flex;margin:0 0 var(--sp-3);padding:3px 9px;border-radius:12px;background:var(--bg-green);color:var(--dot-green);font:600 var(--fs-xs) var(--font-mono)}
+body.fast-plan .cmt-btn,body.fast-plan .comment-section{display:none}
 #sections{display:grid;grid-template-columns:1fr 1fr;gap:var(--sp-5)}
 @media(max-width:720px){#sections{grid-template-columns:1fr}}
 .section{margin-bottom:var(--sp-4)}
@@ -82,7 +84,8 @@ textarea.comment-box:hover:not(:focus){border-color:var(--text-muted)}
 const PLAN_BODY = `
 <div class="main">
   <h1 id="title"></h1>
-  <p class="hint">Each section is rendered markdown — <strong>click any section to edit</strong> (save with blur or ⌘/Ctrl+Enter). Use the 💬 icon to leave a comment on a section, and confirm the <strong>external dependencies</strong> below — new packages, libraries, or services this plan needs, never tasks or todos. When everything looks right click <strong>Accept</strong>; if you edit anything or add a comment the button becomes <strong>Send review</strong> so Claude can revise first.</p>
+  <div class="fast-track" id="fast-track" hidden>Structured plan · ready for review</div>
+  <p class="hint" id="plan-hint">Each section is rendered markdown — <strong>click any section to edit</strong> (save with blur or ⌘/Ctrl+Enter). Use the 💬 icon to leave a comment on a section, and confirm the <strong>external dependencies</strong> below — new packages, libraries, or services this plan needs, never tasks or todos. When everything looks right click <strong>Accept</strong>; if you edit anything or add a comment the button becomes <strong>Send review</strong> so Claude can revise first.</p>
   <div class="deps-panel">
     <div class="shead"><span class="slabel">External dependencies</span><span class="chip-design" id="design-chip"></span></div>
     <div class="dep-chips" id="dep-chips"></div>
@@ -123,6 +126,11 @@ SECTION_KEYS.forEach(k => comments[k] = []);
 let activeKey = null;
 
 document.getElementById('title').textContent = D.title || 'Plan Review';
+if(D.fastTrack){
+  document.body.classList.add('fast-plan');
+  document.getElementById('fast-track').hidden = false;
+  document.getElementById('plan-hint').innerHTML = 'This complete plan skipped discovery and is ready for review. <strong>Click any section to edit it directly</strong>, confirm the external dependencies, then Accept to save it and continue to feature breakdown.';
+}
 const dchip = document.getElementById('design-chip');
 if(D.designFile){ dchip.className = 'chip-design on'; dchip.textContent = 'design params attached'; dchip.title = D.designFile; }
 else { dchip.className = 'chip-design off'; dchip.textContent = 'no design params \\u2014 /iterator-design'; dchip.title = 'UI-touching plans should capture memory/design.md first'; }
@@ -212,6 +220,7 @@ function deleteComment(i){ comments[activeKey].splice(i,1); renderComments(); re
 function collectComments(){ const out=[]; SECTION_KEYS.forEach(k => comments[k].forEach(text => out.push({section:k, text}))); return out; }
 
 function hasChanges(){
+  if(D.fastTrack) return false;
   if(document.getElementById('global-comment').value.trim()) return true;
   if(collectComments().length) return true;
   if(JSON.stringify(sections) !== ORIG_SECTIONS) return true;
@@ -232,9 +241,16 @@ function onPrimary(){
 `;
 
 export function render(data) {
-  return renderPage({
-    step: 'plan', subtitle: '/ plan review', branch: data.branch, title: data.title,
-    data, css: PLAN_CSS, body: PLAN_BODY, clientJs: PLAN_JS,
-    primaryIdle: 'Accept', primaryChanged: 'Send review',
-  });
+	return renderPage({
+		step: "plan",
+		subtitle: "/ plan review",
+		branch: data.branch,
+		title: data.title,
+		data,
+		css: PLAN_CSS,
+		body: PLAN_BODY,
+		clientJs: PLAN_JS,
+		primaryIdle: "Accept",
+		primaryChanged: "Send review",
+	});
 }
diff --git a/lib/views/planning.mjs b/lib/views/planning.mjs
index 7b9c314..6fc8f59 100644
--- a/lib/views/planning.mjs
+++ b/lib/views/planning.mjs
@@ -36,9 +36,11 @@ const PLANNING_CSS = `
 .backlog{background:var(--surface);border:1px solid var(--border);border-radius:var(--radius-card);box-shadow:var(--shadow-card);padding:var(--sp-4);margin-top:var(--sp-6)}
 .backlog-head{display:flex;align-items:center;justify-content:space-between;gap:var(--sp-3);flex-wrap:wrap;margin-bottom:var(--sp-3)}
 .backlog-head h2{font-size:var(--fs-md);font-weight:600}.backlog-head p{font-size:var(--fs-xs);color:var(--text-muted)}
+.backlog-tools{display:flex;align-items:center;gap:var(--sp-2);flex-wrap:wrap}.backlog-tools .filters{display:flex;gap:var(--sp-1)}
+.backlog-tools button.sel{background:var(--accent);border-color:var(--accent);color:var(--accent-fg)}
 .backlog-form{display:grid;grid-template-columns:120px minmax(180px,1fr);gap:var(--sp-2);align-items:start}
 .backlog-form select,.backlog-form input,.backlog-form textarea{width:100%;background:var(--bg);border:1px solid var(--border);border-radius:var(--radius-sm);padding:8px 10px;color:var(--text);font:inherit;font-size:var(--fs-sm)}
-.backlog-form textarea{min-height:64px;resize:vertical;grid-column:1/-1}.backlog-form .btns{grid-column:1/-1}.backlog-list{display:grid;gap:var(--sp-2);margin-top:var(--sp-4)}
+.backlog-form textarea{min-height:64px;resize:vertical;grid-column:1/-1}.backlog-form .at-wrap{position:relative;grid-column:1/-1}.backlog-form .at-wrap textarea{grid-column:auto}.backlog-form .btns{grid-column:1/-1}.backlog-list{display:grid;gap:var(--sp-2);margin-top:var(--sp-4)}
 .backlog-item{border-top:1px solid var(--border);padding-top:var(--sp-3)}.backlog-item:first-child{border-top:0;padding-top:0}
 .backlog-item-main{display:flex;gap:var(--sp-2);align-items:flex-start}.backlog-item-main input{margin-top:4px}.backlog-item-title{font-size:var(--fs-sm);font-weight:600}.backlog-item-details{font-size:var(--fs-xs);color:var(--text-muted);margin-top:2px;white-space:pre-wrap}.backlog-empty{font-size:var(--fs-sm);color:var(--text-muted)}
 .retired{margin-top:var(--sp-6)}.retired .sec-title{margin-top:0}.retired-list{display:grid;gap:var(--sp-3)}
@@ -56,11 +58,12 @@ function clearPlanDraft(){ try { localStorage.removeItem(PLAN_DRAFT_KEY); } catc
 
 async function backlogAction(payload, button, message){
   if(button){ button.disabled = true; button.dataset.label = button.textContent; button.textContent = 'Saving…'; }
-  await post({ type:'backlog', ...payload }, message || 'Saved', { allowWhileWorking:true });
+  const saved = await post({ type:'backlog', ...payload }, message || 'Saved', { allowWhileWorking:true });
   // During an agent turn the server deliberately does not refresh this view:
   // doing so would clear the model-working guard. Restore the local control so
   // more filesystem-only backlog edits remain possible until turn-end refresh.
   if(button){ button.disabled = false; if(button.dataset.label) button.textContent = button.dataset.label; delete button.dataset.label; }
+  return saved;
 }
 function selectedBacklogGoal(){
   const selected = (D.backlog || []).filter(item => item.selected);
@@ -68,6 +71,43 @@ function selectedBacklogGoal(){
   return 'Create a plan from these saved backlog candidates:\\n\\n' + selected.map(item =>
     '[' + item.kind + '] ' + item.title + (item.details ? '\\n' + item.details : '')).join('\\n\\n');
 }
+function parseStructuredPlan(value){
+  const lines = String(value || '').replaceAll('\\r', '').split('\\n');
+  const sections = {goal:[], architecture:[], keyDecisions:[], dependencies:[]};
+  let title = '', active = null, found = 0;
+  const heading = line => {
+    let text = line.trim();
+    const markdown = text.startsWith('#');
+    while(text.startsWith('#')) text = text.slice(1).trim();
+    if(text.endsWith(':')) text = text.slice(0, -1).trim();
+    const compact = text.toLowerCase().replaceAll(' ', '').replaceAll('-', '');
+    const keys = {goal:'goal',architecture:'architecture',keydecisions:'keyDecisions',decisions:'keyDecisions',dependencies:'dependencies',externaldependencies:'dependencies'};
+    return (markdown || line.trim().endsWith(':')) ? (keys[compact] || null) : null;
+  };
+  for(const raw of lines){
+    const trimmed = raw.trim();
+    const key = heading(raw);
+    if(key){ active = key; found++; continue; }
+    if(!title && trimmed.toLowerCase().startsWith('title:')){
+      title = trimmed.slice(6).trim(); continue;
+    }
+    if(!title && trimmed.startsWith('#') && !active){
+      title = trimmed.slice(1).trim(); continue;
+    }
+    if(active) sections[active].push(raw);
+  }
+  const clean = key => sections[key].join('\\n').trim();
+  const goal = clean('goal'), architecture = clean('architecture'), keyDecisions = clean('keyDecisions');
+  if(found < 3 || !goal || !architecture || !keyDecisions) return null;
+  if(!title) title = goal.split('\\n').find(Boolean).slice(0, 80);
+  const dependencies = sections.dependencies.map(line => {
+    let item = line.trim();
+    if(item.startsWith('- ') || item.startsWith('* ')) item = item.slice(2).trim();
+    return item;
+  }).filter(item => item && !['none','n/a','no new dependencies'].includes(item.toLowerCase()));
+  const description = goal.split('\\n').map(line => line.trim()).find(Boolean).slice(0, 160);
+  return {title, description, plan:{goal, architecture, keyDecisions}, dependencies};
+}
 
 // Minimal @-file suggestions on the goal box: substring filter over the
 // gathered tracked-file list, arrow keys + Enter/Tab to accept.
@@ -136,7 +176,7 @@ function render(){
       '<h2>No plan yet</h2><p>iterator keeps its state in a memory/ bundle in your repo.<br>Type a goal and start turning it into a reviewable plan.</p>';
     const goal = document.createElement('textarea');
     goal.className = 'goal';
-    goal.placeholder = 'What are you building and why? (1\\u20133 sentences \\u2014 optional, saves a question round; @ mentions repo files)';
+    goal.placeholder = 'Describe a goal, or paste a complete plan with Goal, Architecture, and Key Decisions headings. @ mentions repo files.';
     // The iframe is recreated on every tab switch/refresh — keep the unsent
     // goal in browser storage, cleared only once a plan actually starts.
     try { goal.value = localStorage.getItem(PLAN_DRAFT_KEY) || ''; } catch(e){}
@@ -172,9 +212,18 @@ function render(){
     }
     const b = document.createElement('button');
     b.className = 'act' + (D.knowledgeInitialized ? ' primary-act' : ''); b.textContent = 'Create plan';
-    b.addEventListener('click', () =>
-      post({ type:'action', action:'plan', feature:null, prompt: goal.value.trim() || null }, 'Starting /iterator-plan')
-        .then(() => { if(__submitted) clearPlanDraft(); }));
+    b.addEventListener('click', () => {
+      const prompt = goal.value.trim() || null;
+      const structuredPlan = parseStructuredPlan(prompt);
+      return post({
+        type:'action',
+        action:structuredPlan ? 'plan-fast-track' : 'plan',
+        feature:null,
+        prompt,
+        structuredPlan
+      }, structuredPlan ? 'Opening plan review' : 'Starting /iterator-plan')
+        .then(() => { if(__submitted && !structuredPlan) clearPlanDraft(); });
+    });
     btns.appendChild(b);
     hero.appendChild(btns);
     w.appendChild(hero);
@@ -183,65 +232,17 @@ function render(){
     return;
   }
 
-  // Plan bar: title, status, progress count, and the stage-driven lifecycle
-  // buttons. Execution controls (auto mode, Test/Implement/Review) live on
-  // the Work surface.
-  const done = (D.progress&&D.progress.done)||0, total = (D.progress&&D.progress.total)||CH.length;
-  const bar = document.createElement('div');
-  bar.className = 'planbar';
-  bar.innerHTML = '<span class="pt">'+esc(D.plan.title||'Plan')+'</span>'+
-    '<span class="pst '+(D.plan.status==='approved'?'approved':'draft')+'">'+esc(D.plan.status||'draft')+'</span>'+
-    (total
-      ? '<span class="pcount">'+done+' / '+total+' features done</span>'
-      : '<span class="pcount">not broken into features yet</span>');
-  const revise = document.createElement('button');
-  revise.className='act'; revise.textContent='Revise plan';
-  revise.addEventListener('click', () => action('plan', null, 'Starting /iterator-plan'));
-  bar.appendChild(revise);
-  // Before featuring, this button IS the continue action — make it read and
-  // look like one instead of a "Re-feature" that has nothing to redo.
-  const refeature = document.createElement('button');
-  refeature.className = D.stage==='needs-features' ? 'act primary-act' : 'act';
-  refeature.textContent = D.stage==='needs-features' ? 'Feature the plan' : 'Re-feature';
-  refeature.title = D.stage==='needs-features'
-    ? 'Next step: break the approved plan into small, dependency-ordered features (/iterator-feature)'
-    : 'Redraw the feature set from the plan (/iterator-feature)';
-  refeature.addEventListener('click', () => action('feature', null, 'Starting /iterator-feature'));
-  bar.appendChild(refeature);
-  // Everything landed → the whole-plan review; every feature done → retire.
-  // Both derive from the server-computed stage, never recomputed here.
-  if(D.stage==='awaiting-plan-review' || D.stage==='retirable'){
-    const rp = document.createElement('button');
-    rp.className = 'act primary-act';
-    rp.textContent = D.plan.planReviewed ? 'Re-review plan' : 'Review plan';
-    rp.title = D.plan.planReviewed
-      ? 'Plan reviewed '+D.plan.planReviewed+' \\u2014 run the whole-plan review again'
-      : 'Review all changes and commits against the plan\\u2019s goals and decisions';
-    rp.addEventListener('click', () => action('review-plan', null, 'Starting /iterator-review-plan'));
-    bar.appendChild(rp);
-  }
-  if(D.stage==='retirable'){
-    const retire = document.createElement('button');
-    retire.className='act primary-act'; retire.textContent='Retire plan';
-    retire.title='Condense the finished plan into a decisions/ memory and archive its features';
-    confirmButton(retire, 'Retires the plan \\u2014 click again', () =>
-      action('retire', null, 'Starting plan retirement'));
-    bar.appendChild(retire);
-  }
-  // Cancel: abandon the plan entirely — archives the bundle side and deletes
-  // the plan branch/worktree, so the armed label spells out what is at stake.
-  const cancelPlan = document.createElement('button');
-  cancelPlan.className = 'act danger';
-  cancelPlan.textContent = 'Cancel plan';
-  const dirtyWarn = (D.dirty && D.dirty.count)
-    ? D.dirty.count + ' uncommitted file' + (D.dirty.count!==1?'s':'') + ' + '
-    : '';
-  cancelPlan.title = 'Abandon this plan: archive it and DELETE its branch/worktree'
-    + (dirtyWarn ? ' \\u2014 \\u26a0 ' + dirtyWarn + 'unmerged commits will be lost' : '');
-  confirmButton(cancelPlan, '\\u26a0 Deletes ' + dirtyWarn + 'branch \\u2014 click again', () =>
-    action('cancel-plan', null, 'Cancelling plan'));
-  bar.appendChild(cancelPlan);
-  w.appendChild(bar);
+  // Planning is a staging surface once a plan exists. Active lifecycle,
+  // progress, and feature controls stay together on Work.
+  const hero = document.createElement('div');
+  hero.className = 'hero';
+  hero.innerHTML = '<h2>Planning is staged</h2><p><strong>'+esc(D.plan.title||'This plan')+'</strong> is active on Work.<br>Use Work for progress, feature actions, and plan lifecycle controls.</p>';
+  const open = document.createElement('button');
+  open.className = 'act primary-act';
+  open.textContent = 'Open Work';
+  open.addEventListener('click', () => action('hub', null, 'Opening Work'));
+  hero.appendChild(open);
+  w.appendChild(hero);
   renderBacklog(w);
   renderRetired(w);
 }
@@ -250,13 +251,15 @@ function render(){
 function renderBacklog(w){
   const items = Array.isArray(D.backlog) ? D.backlog : [];
   const section = document.createElement('section'); section.className = 'backlog backlog-active';
-  section.innerHTML = '<div class="backlog-head"><div><h2>Idea backlog</h2><p>Saved ideas and bugs stay separate from active plan features.</p></div></div>';
+  section.innerHTML = '<div class="backlog-head"><div><h2>Idea backlog</h2><p>Saved ideas and bugs stay separate from active plan features.</p></div><div class="backlog-tools"><div class="filters" aria-label="Filter backlog"><button class="act sel" type="button" data-kind="all">All</button><button class="act" type="button" data-kind="idea">Ideas</button><button class="act" type="button" data-kind="bug">Bugs</button></div><button class="act" type="button" data-bulk="select">Select visible</button><button class="act" type="button" data-bulk="deselect">Deselect visible</button></div></div>';
   const form = document.createElement('form'); form.className = 'backlog-form';
   form.innerHTML = '<select aria-label="Candidate type"><option value="idea">Idea</option><option value="bug">Bug</option></select>'+
     '<input required maxlength="160" placeholder="Short title" aria-label="Backlog title">'+
-    '<textarea maxlength="4000" placeholder="Why it matters, context, or a repro" aria-label="Backlog details"></textarea>'+
+    '<div class="at-wrap"><textarea maxlength="4000" placeholder="Why it matters, context, or a repro" aria-label="Backlog details"></textarea><div class="at-menu"></div></div>'+
     '<div class="btns"><button class="act primary-act" type="submit">Save candidate</button><button class="act" type="button" hidden>Cancel edit</button></div>';
   const [kind, title, details] = form.querySelectorAll('select,input,textarea');
+  const detailsMenu = form.querySelector('.at-menu');
+  wireAtMenu(details, detailsMenu);
   const [save, cancel] = form.querySelectorAll('button');
   let editing = null;
   const reset = () => { editing = null; form.reset(); save.textContent = 'Save candidate'; cancel.hidden = true; };
@@ -267,12 +270,21 @@ function renderBacklog(w){
   cancel.addEventListener('click', reset);
   section.appendChild(form);
   const list = document.createElement('div'); list.className = 'backlog-list bounded-list';
+  const rows = new Map();
+  let filter = 'all';
+  let refreshHandoff = () => {};
   if(!items.length) list.innerHTML = '<p class="backlog-empty">No saved candidates yet.</p>';
   for(const item of items){
-    const row = document.createElement('div'); row.className = 'backlog-item';
+    const row = document.createElement('div'); row.className = 'backlog-item'; row.dataset.kind = item.kind;
+    rows.set(item.id, row);
     row.innerHTML = '<div class="backlog-item-main"><input type="checkbox" aria-label="Select '+esc(item.title)+'" '+(item.selected?'checked':'')+'><div><div class="backlog-item-title">'+esc(item.title)+' <span class="chip cmut">'+esc(item.kind)+'</span></div><div class="backlog-item-details">'+esc(item.details || '')+'</div></div></div>';
     const toggle = row.querySelector('input');
-    toggle.addEventListener('change', () => backlogAction({ action:'select', id:item.id, selected:toggle.checked }, toggle, toggle.checked ? 'Candidate selected' : 'Candidate deselected'));
+    toggle.addEventListener('change', async () => {
+      const selected = toggle.checked;
+      const saved = await backlogAction({ action:'select', id:item.id, selected }, toggle, selected ? 'Candidate selected' : 'Candidate deselected');
+      if(saved){ item.selected = selected; refreshHandoff(); }
+      else toggle.checked = !selected;
+    });
     const buttons = document.createElement('div'); buttons.className = 'btns';
     const edit = document.createElement('button'); edit.className = 'act'; edit.type = 'button'; edit.textContent = 'Edit';
     edit.addEventListener('click', () => { editing = item.id; kind.value = item.kind; title.value = item.title; details.value = item.details || ''; save.textContent = 'Update candidate'; cancel.hidden = false; title.focus(); });
@@ -280,16 +292,43 @@ function renderBacklog(w){
     confirmButton(remove, 'Really delete — click again', () => backlogAction({ action:'delete', id:item.id }, remove, 'Candidate deleted'));
     buttons.append(edit, remove); row.appendChild(buttons); list.appendChild(row);
   }
+  const visibleItems = () => items.filter(item => filter==='all' || item.kind===filter);
+  const applyFilter = kind => {
+    filter = kind;
+    section.querySelectorAll('.filters [data-kind]').forEach(button => button.classList.toggle('sel', button.dataset.kind===kind));
+    rows.forEach((row, id) => { row.hidden = !visibleItems().some(item => item.id===id); });
+  };
+  section.querySelectorAll('.filters [data-kind]').forEach(button =>
+    button.addEventListener('click', () => applyFilter(button.dataset.kind)));
+  const bulkSelect = async (selected, button) => {
+    const targets = visibleItems().filter(item => Boolean(item.selected)!==selected);
+    if(!targets.length) return;
+    const saved = await backlogAction(
+      { action:'select-many', ids:targets.map(item => item.id), selected },
+      button,
+      selected ? 'Visible candidates selected' : 'Visible candidates deselected'
+    );
+    if(!saved) return;
+    for(const item of targets){
+      item.selected = selected;
+      const checkbox = rows.get(item.id).querySelector('input'); checkbox.checked = selected;
+    }
+    refreshHandoff();
+  };
+  section.querySelector('[data-bulk="select"]').addEventListener('click', event => bulkSelect(true, event.currentTarget));
+  section.querySelector('[data-bulk="deselect"]').addEventListener('click', event => bulkSelect(false, event.currentTarget));
   section.appendChild(list);
-  const goal = selectedBacklogGoal();
-  if(goal){
-    const handoff = document.createElement('button'); handoff.className = 'act primary-act'; handoff.type = 'button';
-    handoff.textContent = D.plan ? 'Selected candidates saved' : 'Plan selected candidates';
-    handoff.title = D.plan ? 'Retire or finish the active plan before starting a new one.' : 'Start a new plan with the selected candidates as its initial goal. They are removed from the backlog only after the plan is approved.';
-    handoff.disabled = Boolean(D.plan);
-    handoff.addEventListener('click', () => action('plan', null, 'Starting /iterator-plan from backlog', goal));
-    section.appendChild(handoff);
-  }
+  const handoff = document.createElement('button'); handoff.className = 'act primary-act'; handoff.type = 'button';
+  handoff.textContent = D.plan ? 'Selected candidates saved' : 'Plan selected candidates';
+  handoff.title = D.plan ? 'Retire or finish the active plan before starting a new one.' : 'Start a new plan with the selected candidates as its initial goal. They are removed from the backlog only after the plan is approved.';
+  handoff.disabled = Boolean(D.plan);
+  handoff.addEventListener('click', () => {
+    const goal = selectedBacklogGoal();
+    if(goal) action('plan', null, 'Starting /iterator-plan from backlog', goal);
+  });
+  refreshHandoff = () => { handoff.hidden = !selectedBacklogGoal(); };
+  refreshHandoff();
+  section.appendChild(handoff);
   w.appendChild(section);
 }
 
diff --git a/lib/views/settings.mjs b/lib/views/settings.mjs
index 2050b1c..8ae1d49 100644
--- a/lib/views/settings.mjs
+++ b/lib/views/settings.mjs
@@ -10,10 +10,10 @@
  *             models?: [{id,label}] }               // pi model registry, for model dropdowns
  *   output: one JSON line to stdout —
  *     { type:"settings", values:{<changed key>:<value>} }
- *     plus the shared { type:"cancel" } / { type:"timeout" }.
+ *     | { type:"settings-close" } when the shell-owned modal is dismissed.
  */
-import { renderPage } from '../ui.mjs';
-import { SETTINGS_DEFS } from '../settings.mjs';
+import { renderPage } from "../ui.mjs";
+import { SETTINGS_DEFS } from "../settings.mjs";
 
 const CSS = `
 .main{max-width:760px;margin:0 auto;padding:28px var(--sp-5)}
@@ -48,11 +48,41 @@ const BODY = `
 
 // Render order: behavior first, then models, then the knobs.
 const GROUPS = [
-  ['Flow', ['auto_mode', 'testing_default', 'review_required', 'branch_per_plan', 'worktree_per_plan', 'auto_retire_prompt']],
-  ['Models', ['planner_model', 'planner_thinking', 'implementer_model', 'implementer_thinking',
-    'tester_model', 'tester_thinking', 'reviewer_model', 'reviewer_thinking',
-    'plan_reviewer_model', 'plan_reviewer_thinking']],
-  ['Limits & bookkeeping', ['max_review_iterations', 'block_commit_on_leftovers', 'memorize_nudge', 'usage_ledger']],
+	[
+		"Flow",
+		[
+			"auto_mode",
+			"testing_default",
+			"review_required",
+			"branch_per_plan",
+			"worktree_per_plan",
+			"auto_retire_prompt",
+		],
+	],
+	[
+		"Models",
+		[
+			"planner_model",
+			"planner_thinking",
+			"implementer_model",
+			"implementer_thinking",
+			"tester_model",
+			"tester_thinking",
+			"reviewer_model",
+			"reviewer_thinking",
+			"plan_reviewer_model",
+			"plan_reviewer_thinking",
+		],
+	],
+	[
+		"Limits & bookkeeping",
+		[
+			"max_review_iterations",
+			"block_commit_on_leftovers",
+			"memorize_nudge",
+			"usage_ledger",
+		],
+	],
 ];
 
 const JS = `
@@ -153,15 +183,26 @@ function hasChanges(){
 }
 function onPrimary(){
   const values = changedValues();
-  if(!Object.keys(values).length){ post({ type:'cancel' }, 'Nothing changed'); return; }
+  if(!Object.keys(values).length){
+    post({ type:'settings-close' }, 'Settings closed', { allowWhileWorking:true });
+    return;
+  }
   post({ type:'settings', values }, 'Settings saved');
 }
 `;
 
 export function render(data) {
-  return renderPage({
-    step: 'settings', subtitle: '/ settings', branch: data.branch, title: data.plan,
-    data, css: CSS, body: BODY, clientJs: JS,
-    primaryIdle: 'Close', primaryChanged: 'Save settings',
-  });
+	return renderPage({
+		step: "settings",
+		subtitle: "/ settings",
+		branch: data.branch,
+		title: data.plan,
+		data,
+		css: CSS,
+		body: BODY,
+		clientJs: JS,
+		primaryIdle: "Close",
+		primaryChanged: "Save settings",
+		cancel: false,
+	});
 }
diff --git a/lib/views/widgets.mjs b/lib/views/widgets.mjs
index de8e9d7..18a6280 100644
--- a/lib/views/widgets.mjs
+++ b/lib/views/widgets.mjs
@@ -64,7 +64,7 @@ function action(act, feature, msg, prompt){
 function testBadge(c){
   if(!c.testsStatus || c.testsStatus==='none') return '';
   return c.testsStatus==='green' ? '<span class="chip cg"><i class="sdot"></i>tests green</span>'
-                                 : '<span class="chip cr"><i class="sdot"></i>tests red</span>';
+                                 : '<span class="chip cr"><i class="sdot"></i>tests committed · intentionally red</span>';
 }
 function sizeChip(c){
   const cls = c.size==='large'?'cr':c.size==='medium'?'cy':'cg';
diff --git a/lib/write.mjs b/lib/write.mjs
index 8f04e37..ee837e1 100644
--- a/lib/write.mjs
+++ b/lib/write.mjs
@@ -898,16 +898,28 @@ function writeBacklog(payload, root) {
 			updated: nowIso(),
 		};
 		items[i] = changed;
-	} else if (action === "select") {
-		const id = String(payload.id || "").trim();
-		if (!/^[a-z0-9][a-z0-9-]*$/.test(id))
+	} else if (action === "select" || action === "select-many") {
+		const ids =
+			action === "select"
+				? [String(payload.id || "").trim()]
+				: Array.isArray(payload.ids)
+					? [...new Set(payload.ids.map((id) => String(id).trim()))]
+					: [];
+		if (!ids.length) fail("backlog ids must be a non-empty array");
+		if (ids.some((id) => !/^[a-z0-9][a-z0-9-]*$/.test(id)))
 			fail("backlog id must be a kebab-case slug");
 		if (typeof payload.selected !== "boolean")
 			fail("backlog selected must be a boolean");
-		const i = items.findIndex((item) => item.id === id);
-		if (i === -1) fail(`no backlog item '${id}'`);
-		changed = { ...items[i], selected: payload.selected, updated: nowIso() };
-		items[i] = changed;
+		const missing = ids.filter((id) => !items.some((item) => item.id === id));
+		if (missing.length) fail(`no backlog item '${missing[0]}'`);
+		const changedIds = new Set(ids);
+		const updated = nowIso();
+		changed = [];
+		for (let i = 0; i < items.length; i++) {
+			if (!changedIds.has(items[i].id)) continue;
+			items[i] = { ...items[i], selected: payload.selected, updated };
+			changed.push(items[i]);
+		}
 	} else if (action === "delete") {
 		const id = String(payload.id || "").trim();
 		if (!/^[a-z0-9][a-z0-9-]*$/.test(id))
@@ -917,9 +929,10 @@ function writeBacklog(payload, root) {
 		changed = items[i];
 		items.splice(i, 1);
 	} else {
-		fail("backlog action must be create, edit, select, or delete");
+		fail("backlog action must be create, edit, select, select-many, or delete");
 	}
 
+	const changedItems = Array.isArray(changed) ? changed : [changed];
 	const dir = join(b.memDir, "backlog");
 	mkdirSync(dir, { recursive: true });
 	writeFileSync(backlogPath(b), backlogIndex(items));
@@ -927,9 +940,16 @@ function writeBacklog(payload, root) {
 	prependLog(
 		b.memDir,
 		payload.log ||
-			`**Backlog**: ${action} ${changed.id}${action === "select" ? (changed.selected ? " (selected)" : " (deselected)") : ""}.`,
+			`**Backlog**: ${action} ${changedItems.map((item) => item.id).join(", ")}${action.startsWith("select") ? (payload.selected ? " (selected)" : " (deselected)") : ""}.`,
 	);
-	return { op: "backlog", action, item: changed, items, memoryDir: b.memDir };
+	return {
+		op: "backlog",
+		action,
+		item: changedItems[0],
+		changedItems,
+		items,
+		memoryDir: b.memDir,
+	};
 }
 
 // ---------------------------------------------------------------------------
@@ -3207,8 +3227,9 @@ const SCHEMAS = {
 	},
 	backlog: {
 		op: "backlog",
-		action: "create|edit|select|delete",
-		"id?": "kebab-case item id (required except create)",
+		action: "create|edit|select|select-many|delete",
+		"id?": "kebab-case item id (required for edit/select/delete)",
+		"ids?": ["item ids (required for select-many)"],
 		"title?": "1-160 characters (required for create/edit)",
 		"details?": "up to 4000 characters (required for create/edit)",
 		"kind?": "idea|bug (required for create/edit)",
diff --git a/memory/backlog/index.md b/memory/backlog/index.md
index 12e44c7..4d9a789 100644
--- a/memory/backlog/index.md
+++ b/memory/backlog/index.md
@@ -2,10 +2,12 @@
 type: Backlog
 title: Iterator backlog
 description: Saved ideas and bugs kept separate from active plan features.
-items: "[]"
-timestamp: 2026-07-18T06:52:21.220Z
+items: "[{\"id\":\"red-mode-tests-should-show-the-real-test-code-for-verification\",\"title\":\"red mode tests should show the real test code for verification\",\"details\":\"as tests are important and the real implementation is most important the user needs to verify the code for each test.\",\"kind\":\"idea\",\"selected\":true,\"created\":\"2026-07-20T13:29:32.558Z\",\"updated\":\"2026-07-20T16:04:41.379Z\"},{\"id\":\"red-mode-tests-should-show-the-real-test-code-for-verification-2\",\"title\":\"red mode tests should show the real test code for verification\",\"details\":\"as tests are important and the real implementation is most important the user needs to verify the code for each test.\",\"kind\":\"idea\",\"selected\":true,\"created\":\"2026-07-20T13:29:35.077Z\",\"updated\":\"2026-07-20T16:04:42.566Z\"},{\"id\":\"remove-text-erferences-to-claude\",\"title\":\"remove text erferences to claude\",\"details\":\"Remove tex references to claude just write Agent\",\"kind\":\"idea\",\"selected\":true,\"created\":\"2026-07-20T16:04:39.455Z\",\"updated\":\"2026-07-20T16:04:43.223Z\"}]"
+timestamp: 2026-07-20T16:04:43.223Z
 ---
 
 # Backlog
 
-(empty)
+* [idea] red mode tests should show the real test code for verification — selected
+* [idea] red mode tests should show the real test code for verification — selected
+* [idea] remove text erferences to claude — selected
diff --git a/memory/decisions/focus-feature-execution-and-dashboard-ownership.md b/memory/decisions/focus-feature-execution-and-dashboard-ownership.md
new file mode 100644
index 0000000..4769cae
--- /dev/null
+++ b/memory/decisions/focus-feature-execution-and-dashboard-ownership.md
@@ -0,0 +1,35 @@
+---
+type: Decision
+title: Focus feature execution and dashboard ownership
+description: Implementation now starts in fresh feature sessions, Work owns active plans, Settings preserves context as a shell modal, and red tests are explicit handoffs.
+status: accepted
+date: 2026-07-20
+tags: [workflow, dashboard, sessions, settings, testing]
+files: ["extensions/iterator.js", "lib/pi-tools.mjs", "lib/session-server.mjs", "lib/gather.mjs", "lib/views/hub.mjs", "lib/views/planning.mjs", "lib/views/settings.mjs"]
+timestamp: 2026-07-20T14:47:42.230Z
+---
+
+# Outcome
+
+Iterator starts each manual, ready-wave, and automatic feature implementation in a fresh Pi session seeded only with the named deterministic feature command and bundle contract. Handoff state preserves fixed-wave and auto safety state without copying prior conversation context.
+
+Work is the complete active-plan surface: it owns progress, dependency graph, feature execution, revision, review, retirement, cancellation, and escalation controls. Planning remains the staged future-work surface for plan creation, backlog, and retired-plan browsing. Intentional plan and feature approval transitions land on Work, while ordinary refreshes preserve the selected tab.
+
+Settings is shell-owned modal state. It opens above any dashboard tab, preserves the underlying tab, pending round, and Work overlay, replays on reconnect, and dismisses back to that exact context after save or close. Red tests are a normal pre-implementation checkpoint: Work names their recorded paths, labels them as intentionally failing, and sends the fresh implementer their status and paths so the next action is explicitly to drive them green.
+
+# Key decisions
+
+- Lifecycle/readiness remain derived once by gather and `lib/status.mjs`; browser views render supplied state.
+- Fresh-session routing does not alter review gates, feature-wave snapshots, or commit attribution.
+- Settings writes retain the deterministic writer path; modal results never settle the underlying interactive round.
+- Shared root libraries are synchronized into shipped skill copies, and client-script parsing remains covered.
+
+# Trade-offs
+
+The persistent shell adds modal delivery and reconnect state, but avoids destructive iframe navigation and keeps interactive review ownership intact. The concise ambient handoff includes only red-test paths needed for the immediate implementation target rather than duplicating the full feature contract.
+
+# Retired plan
+
+Condensed from plan "Focus feature execution and dashboard ownership" (4 features, archived under /features/archive/2026-07-20-focus-feature-execution-and-dashboard-ownership/).
+
+Token usage: 5337452 in / 67475 out / 40762368 cache-read / 0 cache-write over 224 turns (per-step breakdown in the archived usage.md).
diff --git a/memory/decisions/index.md b/memory/decisions/index.md
index 62bbb77..cdcf207 100644
--- a/memory/decisions/index.md
+++ b/memory/decisions/index.md
@@ -5,12 +5,15 @@ Durable product and implementation choices agents should preserve.
 * [Apply role models to manual turns and reset stale runtime state](/decisions/manual-role-models-and-runtime-reset.md) - Manual Iterator role commands temporarily select configured models, while approved plans and terminal auto runs reset runtime state deterministically.
 * [Backlog planning and parallel feature waves](/decisions/backlog-planning-and-feature-waves.md) - Keep low-risk backlog editing available during active work while implementing a fixed ready-feature wave and reviewing its commit-backed results together.
 * [Consume selected backlog ideas on plan approval](/decisions/consume-accepted-backlog-ideas.md) - Selected idea or bug candidates leave the backlog only after deterministic plan approval.
+* [Focus feature execution and dashboard ownership](/decisions/focus-feature-execution-and-dashboard-ownership.md) - Implementation now starts in fresh feature sessions, Work owns active plans, Settings preserves context as a shell modal, and red tests are explicit handoffs.
 * [Memory relevance, usage costs, and dashboard recovery](/decisions/memory-relevance-usage-and-dashboard-recovery.md) - Iterator now bounds implementation context, keeps knowledge/retirement changes reviewed, prices usage only from project-owned rates, and maintains an authoritative active-work dashboard state.
 * [Parallel feature waves and consolidated review](/decisions/parallel-feature-waves-and-consolidated-review.md) - The dashboard supports fixed dependency-ready implementation waves and commit-backed multi-feature review without weakening explicit acceptance.
 * [Polish dashboard and multi-agent workflows](/decisions/polish-dashboard-and-multi-agent-workflows.md) - Dashboard polish and workflow refinements that clarify project context, constrain settings to usable models, and support deterministic Claude Code feature execution.
 * [Powerline shows the sandbox-published UI port](/decisions/powerline-shows-sandbox-ui-port.md) - The footer trails a ui:PORT segment resolved from ITERATOR_DISPLAY_PORT, falling back to ~/.pisbx-env because sbx run never sources it into pi's environment.
-* [Return to Work when Settings closes](/decisions/settings-close-returns-to-work.md) - Idle Settings close events restore the refreshed Work hub without changing the settings persistence path.
-* [Review navigation and Work context](/decisions/review-navigation-and-work-context.md) - Keep interactive reviews stable across dashboard navigation, make Work the active feature surface, and simplify review controls without weakening the review gate.
+* [Safely restore configured Iterator role models](/decisions/safe-role-model-restoration.md) - Only successfully switched role models are restored, so failed provider changes cannot corrupt the active session credentials.
+* [Settings modal preserves originating dashboard context](/decisions/settings-close-returns-to-work.md) - Open Settings as a shell-owned modal and dismiss it back to the same tab, pending round, and Work overlay.
+* [Streamline backlog planning and knowledge actions](/decisions/streamline-backlog-planning-and-knowledge-actions.md) - Backlog planning remains deterministic during active work, while supplied plans and knowledge-conflict updates stay behind existing approval gates.
 * [Sync shared libs into droppable skills](/decisions/synced-droppable-skill-libs.md) - Shared code is developed in root lib/ and copied into skill folders so skills work when installed together or copied manually.
 * [Unify Iterator dashboard and feature workflow](/decisions/iterator-dashboard-feature-workflow.md) - Dashboard workflows keep backlog candidates separate from active work until selected candidates are consumed by approved plan creation.
 * [Use an OKF markdown bundle in target repos](/decisions/okf-markdown-bundle.md) - Project memory is stored as markdown plus YAML frontmatter in each target repo instead of in an external database.
+* [Work owns active plan context and lifecycle](/decisions/review-navigation-and-work-context.md) - Keep active-plan progress, execution, and lifecycle controls on Work while Planning is reserved for staged future work and archives.
diff --git a/memory/decisions/manual-role-models-and-runtime-reset.md b/memory/decisions/manual-role-models-and-runtime-reset.md
index c5e0fa9..5e80ffb 100644
--- a/memory/decisions/manual-role-models-and-runtime-reset.md
+++ b/memory/decisions/manual-role-models-and-runtime-reset.md
@@ -5,8 +5,8 @@ description: Manual Iterator role commands temporarily select configured models,
 status: accepted
 date: 2026-07-17
 tags: [models, workflow, runtime-state, pi]
-files: ["lib/write.mjs", "extensions/iterator.js", "lib/pi-tools.mjs", "test/write.test.mjs", "test/pi-tools.test.mjs"]
-timestamp: 2026-07-17T17:51:38.757Z
+files: ["lib/write.mjs", "extensions/iterator.js", "lib/pi-tools.mjs", "test/write.test.mjs", "test/pi-tools.test.mjs", "test/extension-model-lifecycle.test.mjs"]
+timestamp: "2026-07-20T13:33:09.768Z"
 ---
 
 ## Decision
@@ -17,10 +17,6 @@ Approved plan creation resets runtime state to manual and idle, clearing active
 
 ## Consequences
 
-Usage rows now reflect the model that executed a manual role command, and stale auto bookkeeping cannot dispatch work into a newly approved plan. Parser coverage verifies exact command handling and plan-review precedence; writer coverage verifies approved resets and draft preservation. The whole-plan review noted that extension lifecycle paths still lack direct hook-level tests.
+A role switch only arms restoration after `pi.setModel` succeeds. Failed or unavailable configured models leave the current model and its credentials untouched; `active` never calls `pi.setModel`. Manual role ownership is FIFO across overlapping starts and ends, so each role input is consumed by one agent turn and only that successful turn can restore the prior model.
 
-# Retired plan
-
-Condensed from plan "Apply role models on manual turns and reset stale auto state" (2 features, archived under /features/archive/2026-07-17-manual-role-models-and-runtime-reset/).
-
-Token usage: 2207627 in / 20868 out / 18306048 cache-read / 0 cache-write over 103 turns (per-step breakdown in the archived usage.md).
+Extension lifecycle tests cover failed tester-to-implementer handoffs, one-time restoration after a successful override, and `active` no-switch behavior. Parser coverage verifies exact command handling and plan-review precedence; writer coverage verifies approved resets and draft preservation.
diff --git a/memory/decisions/review-navigation-and-work-context.md b/memory/decisions/review-navigation-and-work-context.md
index 98b5659..7c334f0 100644
--- a/memory/decisions/review-navigation-and-work-context.md
+++ b/memory/decisions/review-navigation-and-work-context.md
@@ -1,28 +1,24 @@
 ---
 type: Decision
-title: Review navigation and Work context
-description: Keep interactive reviews stable across dashboard navigation, make Work the active feature surface, and simplify review controls without weakening the review gate.
+title: Work owns active plan context and lifecycle
+description: Keep active-plan progress, execution, and lifecycle controls on Work while Planning is reserved for staged future work and archives.
 status: accepted
-date: 2026-07-17
+date: 2026-07-20
 tags: [workflow, review, dashboard, navigation]
-files: ["lib/session-server.mjs", "lib/ui.mjs", "lib/views/hub.mjs", "lib/views/planning.mjs", "lib/views/graph.mjs", "lib/views/review.mjs", "test/session-server.test.mjs", "test/ui.test.mjs", "test/client-js-parse.test.mjs"]
-timestamp: 2026-07-17T16:38:22.809Z
+files: ["lib/session-server.mjs", "lib/ui.mjs", "lib/views/hub.mjs", "lib/views/planning.mjs", "lib/views/graph.mjs", "lib/views/review.mjs", "test/session-server.test.mjs", "test/ui.test.mjs", "test/client-js-parse.test.mjs", "extensions/iterator.js"]
+timestamp: "2026-07-20T14:19:10.566Z"
 ---
 
 ## Decision
 
 The persistent session shell, not its replaceable iframe views, owns unload cancellation. Switching between Planning and Work must preserve a pending review round and its single eventual result; an explicit cancel still pre-empts any pending reload-grace timer.
 
-Work is the home for an active plan's progress, dependency graph, feature cards, execution controls, and feature cancellation. Planning remains focused on backlog collection and plan lifecycle actions. Both surfaces render the same server-derived gather data and never reconstruct readiness or lifecycle state locally.
+Work is the home for every active-plan concern: progress, dependency graph, feature cards, execution controls, revise/re-feature, whole-plan review, retirement, plan cancellation, and feature cancellation. Planning is the staging surface for future plans, backlog collection, and retired-plan browsing; when a plan is active it directs users to Work instead of duplicating lifecycle controls. Both surfaces render the same server-derived gather data and never reconstruct readiness or lifecycle state locally.
+
+Approved plans and accepted feature sets intentionally activate Work. Ordinary refreshes preserve the user's selected tab, so dashboard navigation and pending rounds remain stable.
 
 The review sidebar and detail headings wrap long feature names rather than clipping them. The obsolete lower-right Feedback panel and JSON-preview bookkeeping are removed; feature verdicts, notes, and line comments continue to drive the existing header submission flow.
 
 ## Constraints
 
 Dashboard navigation must not weaken the single-model-flow guard or backlog-write allowance. Changes to root `lib/` are synchronized into shipped skill copies, and inline view scripts retain parse coverage.
-
-# Retired plan
-
-Condensed from plan "Keep active review stable and clarify Planning versus Work" (3 features, archived under /features/archive/2026-07-17-review-navigation-and-work-context/).
-
-Token usage: 817664 in / 25689 out / 13951488 cache-read / 0 cache-write over 113 turns (per-step breakdown in the archived usage.md).
diff --git a/memory/decisions/safe-role-model-restoration.md b/memory/decisions/safe-role-model-restoration.md
new file mode 100644
index 0000000..a4c0816
--- /dev/null
+++ b/memory/decisions/safe-role-model-restoration.md
@@ -0,0 +1,28 @@
+---
+type: Decision
+title: Safely restore configured Iterator role models
+description: Only successfully switched role models are restored, so failed provider changes cannot corrupt the active session credentials.
+status: accepted
+date: 2026-07-20
+tags: [models, credentials, workflow, regression]
+files: ["extensions/iterator.js", "test/extension-model-lifecycle.test.mjs"]
+timestamp: 2026-07-20T13:37:19.997Z
+---
+
+## Outcome
+
+Iterator now treats manual role-model selection as a turn-scoped operation. A configured tester or implementer override that cannot be used leaves the active session model untouched, preventing a red-test-to-implementation handoff from falling onto an invalid provider credential path.
+
+## Decision
+
+Arm restoration only after `pi.setModel` succeeds. The `active` setting never invokes model switching, and each input role is consumed by exactly one agent turn. Track successful manual role changes in FIFO lifecycle order so an overlapping follow-up cannot restore another turn's model.
+
+## Verification
+
+The extension lifecycle suite covers failed override handoff, one-time restoration after a successful override, and active-model no-switch behavior. The full test suite remains green; the lifecycle cases skip only in environments without the optional `typebox` peer dependency.
+
+# Retired plan
+
+Condensed from plan "Fix post-test role model credential corruption" (1 features, archived under /features/archive/2026-07-20-safe-role-model-restoration/).
+
+Token usage: 603657 in / 16114 out / 5499904 cache-read / 0 cache-write over 67 turns (per-step breakdown in the archived usage.md).
diff --git a/memory/decisions/settings-close-returns-to-work.md b/memory/decisions/settings-close-returns-to-work.md
index 30b1d5f..8479fd5 100644
--- a/memory/decisions/settings-close-returns-to-work.md
+++ b/memory/decisions/settings-close-returns-to-work.md
@@ -1,28 +1,24 @@
 ---
 type: Decision
-title: Return to Work when Settings closes
-description: Idle Settings close events restore the refreshed Work hub without changing the settings persistence path.
+title: Settings modal preserves originating dashboard context
+description: Open Settings as a shell-owned modal and dismiss it back to the same tab, pending round, and Work overlay.
 status: accepted
-date: 2026-07-14
-tags: [dashboard, settings, navigation, pi]
-files: ["extensions/iterator.js", "lib/views/settings.mjs", "skills/iterator/lib/views/settings.mjs", "test/session-server.test.mjs", "test/sync.test.mjs"]
-timestamp: 2026-07-14T11:09:27.354Z
+date: 2026-07-20
+tags: [dashboard, settings, navigation, pi, modal]
+files: ["extensions/iterator.js", "lib/session-server.mjs", "lib/views/settings.mjs", "skills/iterator/lib/views/settings.mjs", "test/session-server.test.mjs", "test/ui.test.mjs"]
+timestamp: "2026-07-20T14:35:25.921Z"
 ---
 
 # Decision
 
-Treat Settings as an idle dashboard page. Its Close action emits the existing `cancel` result, which the extension handles by refreshing the Work hub so the user returns to a usable dashboard instead of remaining on Settings.
+Treat Settings as shell-owned modal state, rather than a dashboard tab or a Work iframe document. The gear and `/iterator-settings` open it above whichever tab is active without replacing stored tab HTML, cancelling a pending round, or clearing the Work ownership overlay.
+
+Closing without changes and saving successfully dismiss the modal and reveal the exact originating tab and round. Modal settings results route through the existing deterministic settings writer, but do not settle a pending plan or review. A Settings close remains available while Work is blocked; changing values remains read-only until the agent completes.
 
 # Rationale
 
-Settings changes already use the deterministic writer path. Closing without changes is navigation, not a workflow cancellation, so it must restore the Work view without introducing a separate protocol.
+Settings is global project configuration, not workflow navigation. Preserving the shell context removes the forced jump to Work and lets users inspect or close Settings safely from Planning, Knowledge, Usage, and interactive rounds.
 
 # Consequences
 
-Keep Settings save handling on the existing `settings` result path. Cover idle Settings-close dispatch at the session-server boundary, and route close results through the extension's dashboard refresh behavior.
-
-# Retired plan
-
-Condensed from plan "Restore Work tab after closing Settings" (1 feature, archived under /features/archive/2026-07-14-settings-close-returns-to-work/).
-
-Token usage: 206353 in / 7199 out / 1748992 cache-read / 0 cache-write over 53 turns (per-step breakdown in the archived usage.md).
+The session server owns modal HTML, delivery, reconnect replay, Escape/close behavior, and focus restoration. The extension closes the modal after successful persistence and refreshes only the status strip, preserving the underlying view. Tests cover tab preservation, pending-round safety, blocked Work dismissal, responsive modal sizing, and reconnect replay.
diff --git a/memory/decisions/streamline-backlog-planning-and-knowledge-actions.md b/memory/decisions/streamline-backlog-planning-and-knowledge-actions.md
new file mode 100644
index 0000000..63cf600
--- /dev/null
+++ b/memory/decisions/streamline-backlog-planning-and-knowledge-actions.md
@@ -0,0 +1,32 @@
+---
+type: Decision
+title: Streamline backlog planning and knowledge actions
+description: Backlog planning remains deterministic during active work, while supplied plans and knowledge-conflict updates stay behind existing approval gates.
+status: accepted
+date: 2026-07-20
+tags: [backlog, planning, knowledge, workflow, dashboard]
+files: ["lib/views/planning.mjs", "test/ui.test.mjs", "test/client-js-parse.test.mjs", "lib/session-server.mjs", "extensions/iterator.js", "test/session-server.test.mjs", "lib/write.mjs", "test/write.test.mjs", "lib/views/plan.mjs", "lib/views/memory-review.mjs", "lib/ui.mjs", "lib/views/hub.mjs", "lib/gather.mjs", "test/gather.test.mjs"]
+timestamp: 2026-07-20T16:04:13.795Z
+---
+
+## Outcome
+
+Planning now supports repository-anchored backlog ideas, type filtering, and durable visible-set selection. Those filesystem-backed backlog actions remain available while Work has an owned active-flow overlay, without starting or unblocking another model turn.
+
+A structured full plan is accepted as a prefilled, editable planning draft: it skips redundant discovery only, then still passes through browser approval, the deterministic plan writer, approval-bound candidate consumption, and normal feature slicing. Knowledge reviewers can focus on changed proposal sections, and Work conflicts can initiate a reviewed decision-update request followed by an anchored recheck.
+
+## Decisions
+
+- Keep backlog interaction, lifecycle readiness, and conflict state in their existing server-derived and deterministic writer paths; views render supplied state rather than inferring it.
+- Preserve the explicit approval boundaries for plan creation, feature work, and knowledge updates. Backlog selection remains intent until plan approval; conflict resolution proposes a memory change instead of silently editing a decision.
+- Reuse the persistent session shell so allowed backlog writes refresh Planning without clearing the Work overlay or weakening the single-model-flow guard.
+
+## Trade-offs
+
+The faster full-plan path deliberately retains an editable review and feature-breakdown handoff rather than treating a supplied plan as executable. Highlighting memory changes adds comparison logic, but keeps unsafe raw HTML and acceptance behavior out of the review surface. Shared canonical libraries and their droppable skill copies remain synchronized, with client-script parsing and responsive controls protected by tests.
+
+# Retired plan
+
+Condensed from plan "Streamline backlog planning and knowledge actions" (6 features, archived under /features/archive/2026-07-20-streamline-backlog-planning-and-knowledge-actions/).
+
+Token usage: 3689661 in / 61838 out / 59459584 cache-read / 0 cache-write over 291 turns (per-step breakdown in the archived usage.md).
diff --git a/memory/features/archive/2026-07-20-focus-feature-execution-and-dashboard-ownership/active-plan-workspace.md b/memory/features/archive/2026-07-20-focus-feature-execution-and-dashboard-ownership/active-plan-workspace.md
new file mode 100644
index 0000000..f9312ec
--- /dev/null
+++ b/memory/features/archive/2026-07-20-focus-feature-execution-and-dashboard-ownership/active-plan-workspace.md
@@ -0,0 +1,58 @@
+---
+type: Feature
+title: Active plan workspace
+description: Make Work the complete home for an active plan and land there after plan or feature approval, while Planning stays focused on future work and archives.
+status: done
+size: medium
+depends_on: [fresh-implementation-session]
+files: ["lib/views/hub.mjs", "lib/views/planning.mjs", "extensions/iterator.js", "test/ui.test.mjs", "test/client-js-parse.test.mjs", "test/session-server.test.mjs"]
+memories: [pitfalls/cancel-now-after-grace-timer, pitfalls/client-js-template-literal-escaping, architecture/package-and-skill-layout, architecture/workflow-state-ownership, patterns/safe-browser-rendering, decisions/backlog-planning-and-feature-waves, decisions/consume-accepted-backlog-ideas, decisions/iterator-dashboard-feature-workflow]
+conflicts: "[{\"decision\":\"decisions/review-navigation-and-work-context\",\"note\":\"The accepted plan intentionally moves plan lifecycle controls from Planning to Work, revising the recorded split where Planning owned those actions.\"}]"
+timestamp: "2026-07-20T14:24:04.323Z"
+tags: []
+commits:
+  - sha: c5f6915bf82dc6914b8a02a707422a0c67b1a26a
+    kind: implement
+    date: 2026-07-20
+  - sha: 6bcce8b8e1097d9135c7644c2cb44edc719841f7
+    kind: implement
+    date: 2026-07-20
+reviewed: 2026-07-20
+done: 2026-07-20
+---
+
+# Implementation notes
+
+Move active-plan lifecycle controls—revise, feature/re-feature, whole-plan review, retirement, plan cancellation—and their stage-driven affordances from `planning.mjs` into the Work plan bar beside progress and execution. Planning should continue to show plan creation when none exists, backlog CRUD/selection, and retired plans; with an active plan it should explain that management is on Work rather than duplicate state. Add intentional Work activation after approved plan creation and accepted feature breakdown without changing ordinary refreshes that preserve the user's current tab. Continue rendering server-derived `stage`, readiness, progress, and dirty state; do not rederive lifecycle rules in either view. Follow `memory/design.md`, keep mobile controls usable, and retain inline-client parse tests.
+
+# Snippets
+
+```js
+// Current Planning surface owns active lifecycle buttons.
+revise.addEventListener('click', () => action('plan', null));
+refeature.addEventListener('click', () => action('feature', null));
+```
+
+```js
+const landOnPlanning = preferPlanning && !hub.plan;
+session.showView({ step:'planning', activate: landOnPlanning, ... });
+session.showView({ step:'hub', render: () => VIEWS.hub(hub) });
+```
+
+# Depends on
+
+* [Fresh implementation session](/features/fresh-implementation-session.md)
+
+# Blast radius
+
+Planning/Work navigation, all active plan lifecycle actions, and post-approval landing behavior.
+
+# Decision conflicts
+
+* [decisions/review-navigation-and-work-context](/decisions/review-navigation-and-work-context.md) — The accepted plan intentionally moves plan lifecycle controls from Planning to Work, revising the recorded split where Planning owned those actions.
+
+# Review
+
+## 2026-07-20
+* **Approved** _(agent review: openai-codex/gpt-5.6-sol)_ — Work now consistently owns active-plan progress and lifecycle controls, Planning is limited to staging/backlog/archive concerns, approval transitions activate Work, and the ownership and activation contracts are covered by passing tests.
+* **Needs changes** _(agent review: openai/gpt-5.2)_ — Update lib/views/hub.mjs’s module contract and plan-bar comments (and the synced skill copy) to reflect that Work now owns plan lifecycle actions; they currently still state that Planning owns those controls and omit plan/feature/review-plan/retire/cancel-plan from the output contract. Add extension-level coverage proving plan approval and accepted feature breakdown call refreshHub with activateWork:true, rather than only testing session.showView’s generic activate option.
diff --git a/memory/features/archive/2026-07-20-focus-feature-execution-and-dashboard-ownership/fresh-implementation-session.md b/memory/features/archive/2026-07-20-focus-feature-execution-and-dashboard-ownership/fresh-implementation-session.md
new file mode 100644
index 0000000..0539f46
--- /dev/null
+++ b/memory/features/archive/2026-07-20-focus-feature-execution-and-dashboard-ownership/fresh-implementation-session.md
@@ -0,0 +1,52 @@
+---
+type: Feature
+title: Fresh implementation session
+description: Start every manual, ready-wave, and automatic feature implementation in a new Pi session seeded only with the named feature command and deterministic bundle context.
+status: done
+size: large
+depends_on: []
+files: ["extensions/iterator.js", "lib/pi-tools.mjs", "skills/iterator-implement/SKILL.md", "test/pi-tools.test.mjs", "test/extension-fresh-session.test.mjs"]
+memories: [architecture/package-and-skill-layout, decisions/backlog-planning-and-feature-waves, decisions/iterator-dashboard-feature-workflow, decisions/manual-role-models-and-runtime-reset, decisions/memory-relevance-usage-and-dashboard-recovery, decisions/parallel-feature-waves-and-consolidated-review, decisions/polish-dashboard-and-multi-agent-workflows, decisions/powerline-shows-sandbox-ui-port]
+timestamp: "2026-07-20T14:06:39.909Z"
+tags: []
+commits:
+  - sha: e7a69a4ee3dd962817c766d39109012284a0e530
+    kind: implement
+    date: 2026-07-20
+  - sha: 382ac72398fb7443fe5b9a594105e0704a76a8bb
+    kind: implement
+    date: 2026-07-20
+reviewed: 2026-07-20
+done: 2026-07-20
+---
+
+# Implementation notes
+
+Add one extension-command handoff seam that owns `ExtensionCommandContext.newSession()`. Route explicit `/iterator-implement`, dashboard implementation actions, ready-wave items, and auto-mode implementation decisions through it. Capture only plain strings/state before replacement; inside `withSession`, use only the replacement context to send `/skill:iterator-implement <slug>` with `--auto` or user guidance as needed. Do not append prior conversation. Preserve parent-session lineage if available. Distinguish intentional `session_start.reason === 'new'` implementation handoffs from interrupted auto runs so the extension does not pause them, and rebuild any in-memory wave ownership from durable/safely serialized state or a replacement-safe command. Keep model-role switching and explicit review acceptance unchanged. Update the implement runbook to describe the Pi-only fresh-session boundary while leaving Claude Code deterministic gather/write behavior intact.
+
+# Snippets
+
+```ts
+const result = await ctx.newSession({
+  parentSession: ctx.sessionManager.getSessionFile(),
+  withSession: async (replacementCtx) => {
+    await replacementCtx.sendUserMessage('/skill:iterator-implement auth --auto');
+  },
+});
+```
+
+```js
+// Existing auto/wave paths dispatch implementation as another turn.
+dispatch(action.cmd);
+dispatch(`/skill:iterator-implement ${feature} --auto`);
+```
+
+# Blast radius
+
+Every Pi implementation entrypoint and session lifecycle; mistakes can lose auto/wave progress, duplicate a feature turn, or carry stale session-bound objects into the replacement runtime.
+
+# Review
+
+## 2026-07-20
+* **Approved** _(agent review: openai-codex/gpt-5.6-sol)_ — Approved: replacement sessions now restore the prior step model before switching, apply the implementer role in the fresh turn, preserve ready-wave and autoSteps state, and retain interruption safety on reload; targeted lifecycle helpers and the full suite cover the corrected paths.
+* **Needs changes** _(agent review: openai-codex/gpt-5.6-sol)_ — Fresh-session auto/wave implementations no longer receive the configured implementer role: the old driver now skips applyRole, while the replacement turn's before_agent_start still refuses pendingRole whenever state.mode is auto or featureWave is set. Restore any prior step role before ctx.newSession, mark the intentional handoff in the new extension instance, and apply/restore the implementer role exactly once there. Also carry autoSteps through the handoff so replacing the session for every feature cannot reset and bypass AUTO_MAX_STEPS. Add lifecycle assertions for auto and ready-wave handoffs (the current extension test is skipped when typebox is absent).
diff --git a/memory/features/archive/2026-07-20-focus-feature-execution-and-dashboard-ownership/index.md b/memory/features/archive/2026-07-20-focus-feature-execution-and-dashboard-ownership/index.md
new file mode 100644
index 0000000..d0d62d7
--- /dev/null
+++ b/memory/features/archive/2026-07-20-focus-feature-execution-and-dashboard-ownership/index.md
@@ -0,0 +1,6 @@
+# Features
+
+* [Fresh implementation session](fresh-implementation-session.md) - ✅ done · large · Start every manual, ready-wave, and automatic feature implementation in a new Pi session seeded only with the named feature command and deterministic bundle context.
+* [Active plan workspace](active-plan-workspace.md) - ✅ done · medium · depends: fresh-implementation-session · Make Work the complete home for an active plan and land there after plan or feature approval, while Planning stays focused on future work and archives.
+* [Settings dashboard modal](settings-dashboard-modal.md) - ✅ done · medium · depends: active-plan-workspace · Open project Settings as a modal above any dashboard tab and return to the exact originating context on save or close.
+* [Visible red-test handoff](visible-red-test-handoff.md) - ✅ done · medium · depends: active-plan-workspace · Show committed red tests as the implementation target in Work and carry their exact status and paths into the fresh implementer context.
diff --git a/memory/features/archive/2026-07-20-focus-feature-execution-and-dashboard-ownership/plan.md b/memory/features/archive/2026-07-20-focus-feature-execution-and-dashboard-ownership/plan.md
new file mode 100644
index 0000000..b21f3c1
--- /dev/null
+++ b/memory/features/archive/2026-07-20-focus-feature-execution-and-dashboard-ownership/plan.md
@@ -0,0 +1,53 @@
+---
+type: Plan
+title: Focus feature execution and dashboard ownership
+description: Start each implementation in a minimal fresh session and make Work the single active-plan surface with clear red-test and modal settings state.
+status: approved
+branch: iterator/safe-role-model-handoff
+created: 2026-07-20
+timestamp: "2026-07-20T14:47:15.912Z"
+plan_reviewed: 2026-07-20
+---
+
+# Goal
+
+Make every manual or automatic feature implementation start with a clean Pi session containing only the deterministic feature contract and relevant project knowledge, while simplifying the dashboard so Work owns all active-plan information, Settings opens reliably as a modal from any tab, and committed red tests are unmistakable before implementation.
+
+# Architecture
+
+- Build on `architecture/package-and-skill-layout` and Pi's documented session-replacement lifecycle: route manual, ready-wave, and auto implementation dispatch through an extension command that uses `ExtensionCommandContext.newSession()`, then send only `/skill:iterator-implement <slug>` plus required mode/guidance in the replacement-session context. The implement gather contract, relevant memories, and recorded test paths remain the source of necessary context; prior conversation history is not copied.
+- Preserve `architecture/workflow-state-ownership`: durable plan/feature/auto state remains in the bundle and `lib/status.mjs`/gather payloads. Intentional implementation-session replacement must resume the named feature without being mistaken for an interrupted auto run, losing a fixed-wave item, or bypassing dependency and review gates.
+- Consolidate active-plan presentation in `lib/views/hub.mjs`: progress, dependency graph, feature actions, revise/re-feature/review-plan/retire/cancel controls, and escalation all live on Work. `lib/views/planning.mjs` becomes the future-work surface for plan creation, selected backlog candidates, and retired-plan browsing when no active work needs management.
+- Let `lib/session-server.mjs` own navigation and modal presentation: approved plan/feature transitions intentionally activate Work, while Settings renders in a shell-level modal above whichever tab is open, preserves the originating tab and any pending round, and closes back to that exact context.
+- Make red-test state explicit across gather, Work cards, and implementation kickoff: after `/iterator-test`, show that failing tests are committed and expected, expose the recorded test paths/status to the implementer, and label the implementation action as the green step rather than as a generic pending feature.
+- Keep all UI changes within `memory/design.md` (compact dark control plane, semantic status colors, responsive controls), retain inline-script parse coverage for `pitfalls/client-js-template-literal-escaping`, and run `npm run sync` after canonical `lib/` changes.
+
+# Dependencies
+
+(none)
+
+# Key decisions
+
+- Use a real fresh Pi session for each feature implementation, not compaction or a conversation summary. Record the previous session as lineage if useful, but inject no old chat; deterministic gather data is the complete handoff.
+- Fresh-session routing applies to explicit manual implementation, fixed ready waves, and auto-mode implementation steps. Testing and review stay in their existing turns unless they independently dispatch a new feature implementation.
+- Explicitly revise `decisions/review-navigation-and-work-context`: active plan lifecycle controls move from Planning to Work so Planning contains only future/backlog/archive concerns. This accepted deviation must be captured by `/iterator-memorize` after implementation.
+- Explicitly revise `decisions/settings-close-returns-to-work`: Settings becomes a tab-independent modal and closes to its originating tab/context instead of always forcing Work. This accepted deviation must be captured by `/iterator-memorize` after implementation.
+- Keep `decisions/backlog-planning-and-feature-waves` and `decisions/parallel-feature-waves-and-consolidated-review`: fresh sessions must not change the immutable ready-wave snapshot, single-model-flow guard, commit attribution, or explicit review acceptance.
+- Red remains an expected pre-implementation state: never weaken or auto-fix those tests during test creation; the fresh implementer receives their paths and is responsible for driving them green.
+
+# Features
+
+* [Fresh implementation session](/features/fresh-implementation-session.md) - Start every manual, ready-wave, and automatic feature implementation in a new Pi session seeded only with the named feature command and deterministic bundle context.
+* [Active plan workspace](/features/active-plan-workspace.md) - Make Work the complete home for an active plan and land there after plan or feature approval, while Planning stays focused on future work and archives.
+* [Settings dashboard modal](/features/settings-dashboard-modal.md) - Open project Settings as a modal above any dashboard tab and return to the exact originating context on save or close.
+* [Visible red-test handoff](/features/visible-red-test-handoff.md) - Show committed red tests as the implementation target in Work and carry their exact status and paths into the fresh implementer context.
+
+# Plan review
+
+## 2026-07-20 _(agent review)_
+
+## Whole-plan review
+
+**Clean bill.** The six feature commits cover the goal end to end: implementation dispatch now uses isolated replacement sessions with preserved auto/wave state; Work owns active-plan progress, lifecycle controls, graph, and execution while Planning remains staging/backlog/archive; Settings is shell-owned modal state that preserves the originating tab, pending round, and Work overlay; and committed red tests are explicit Work targets with their paths in the implementation handoff.
+
+The implementation follows the planned architecture: lifecycle/readiness remain gather-derived, modal state lives in `lib/session-server.mjs`, canonical `lib/` changes are synchronized into skill copies, and UI coverage includes inline-script parsing plus responsive modal behavior. Both recorded decision revisions are reflected in the knowledge bundle. No plan-commit scope drift or unaccepted feature remains. The current working-tree dirt is outside the commit-backed plan review and is not a retirement blocker; plan retirement will reset the stale escalated runtime state.
diff --git a/memory/features/archive/2026-07-20-focus-feature-execution-and-dashboard-ownership/settings-dashboard-modal.md b/memory/features/archive/2026-07-20-focus-feature-execution-and-dashboard-ownership/settings-dashboard-modal.md
new file mode 100644
index 0000000..81af5b6
--- /dev/null
+++ b/memory/features/archive/2026-07-20-focus-feature-execution-and-dashboard-ownership/settings-dashboard-modal.md
@@ -0,0 +1,55 @@
+---
+type: Feature
+title: Settings dashboard modal
+description: Open project Settings as a modal above any dashboard tab and return to the exact originating context on save or close.
+status: done
+size: medium
+depends_on: [active-plan-workspace]
+files: ["lib/session-server.mjs", "lib/views/settings.mjs", "extensions/iterator.js", "test/session-server.test.mjs", "test/ui.test.mjs"]
+memories: [pitfalls/cancel-now-after-grace-timer, architecture/browser-server-contract, architecture/package-and-skill-layout, decisions/backlog-planning-and-feature-waves, decisions/iterator-dashboard-feature-workflow, decisions/manual-role-models-and-runtime-reset, decisions/memory-relevance-usage-and-dashboard-recovery, decisions/parallel-feature-waves-and-consolidated-review]
+conflicts: "[{\"decision\":\"decisions/settings-close-returns-to-work\",\"note\":\"The accepted plan replaces forced Work restoration with modal dismissal back to whichever tab and round opened Settings.\"}]"
+timestamp: "2026-07-20T14:36:25.076Z"
+tags: []
+commits:
+  - sha: 791f0a75c5cae0a78ca6839fa63f131d293597f5
+    kind: implement
+    date: 2026-07-20
+done: 2026-07-20
+reviewed: 2026-07-20
+---
+
+# Implementation notes
+
+Promote Settings from a Work-tab iframe document to shell-owned modal state. The gear and `/iterator-settings` must display it immediately regardless of the active tab, without replacing stored tab HTML, cancelling a pending plan/review round, or clearing the Work ownership overlay. Serve/render the existing settings form inside a modal frame or dedicated modal endpoint under `session-server`; post saves through the existing deterministic settings callback. Close and successful save should dismiss the modal and reveal the same originating tab and round. Cover navigation from Planning, Work, Knowledge, and Usage, including while Work is blocked, responsive sizing, focus/close behavior, and reconnect replay. Keep settings persistence unchanged and sync root libraries.
+
+# Snippets
+
+```js
+// Current behavior maps settings into Work and can leave other tabs unchanged.
+if (["planning", "plan", "feature", "archive"].includes(step)) return "planning";
+return "work";
+```
+
+```js
+session.showView({
+  step: 'settings',
+  render: () => VIEWS.settings(payload),
+});
+```
+
+# Depends on
+
+* [Active plan workspace](/features/active-plan-workspace.md)
+
+# Blast radius
+
+Persistent dashboard shell navigation, pending interactive rounds, Work overlay ownership, and every Settings open/save/close path.
+
+# Decision conflicts
+
+* [decisions/settings-close-returns-to-work](/decisions/settings-close-returns-to-work.md) — The accepted plan replaces forced Work restoration with modal dismissal back to whichever tab and round opened Settings.
+
+# Review
+
+## 2026-07-20
+* **Approved** — Accepted through the Settings modal review.
diff --git a/memory/features/archive/2026-07-20-focus-feature-execution-and-dashboard-ownership/usage.md b/memory/features/archive/2026-07-20-focus-feature-execution-and-dashboard-ownership/usage.md
new file mode 100644
index 0000000..26fd347
--- /dev/null
+++ b/memory/features/archive/2026-07-20-focus-feature-execution-and-dashboard-ownership/usage.md
@@ -0,0 +1,48 @@
+---
+type: Usage
+title: Token usage
+description: Per-step model/token ledger and optional project-owned pricing for the active plan — written only by the usage op.
+totals: "{\"steps\":{\"hub\":{\"openai-codex/gpt-5.6-terra\":{\"input\":243028,\"output\":842,\"cacheRead\":811520,\"cacheWrite\":0,\"turns\":9}},\"plan\":{\"openai-codex/gpt-5.6-sol\":{\"input\":118306,\"output\":8575,\"cacheRead\":3610112,\"cacheWrite\":0,\"turns\":21}},\"implement\":{\"openai-codex/gpt-5.6-terra\":{\"input\":2518150,\"output\":40901,\"cacheRead\":22285824,\"cacheWrite\":0,\"turns\":112},\"openai-codex/gpt-5.6-sol\":{\"input\":1332608,\"output\":10554,\"cacheRead\":7464960,\"cacheWrite\":0,\"turns\":42}},\"review\":{\"openai-codex/gpt-5.6-sol\":{\"input\":728855,\"output\":5624,\"cacheRead\":4130304,\"cacheWrite\":0,\"turns\":25},\"openai-codex/gpt-5.6-terra\":{\"input\":396505,\"output\":979,\"cacheRead\":2459648,\"cacheWrite\":0,\"turns\":15}}},\"features\":{\"retire-plan\":{\"input\":243028,\"output\":842,\"cacheRead\":811520,\"cacheWrite\":0,\"turns\":9},\"fresh-implementation-session\":{\"input\":2916375,\"output\":23531,\"cacheRead\":16707072,\"cacheWrite\":0,\"turns\":72},\"active-plan-workspace\":{\"input\":599268,\"output\":13091,\"cacheRead\":8385024,\"cacheWrite\":0,\"turns\":48},\"settings-dashboard-modal\":{\"input\":569517,\"output\":11997,\"cacheRead\":4862464,\"cacheWrite\":0,\"turns\":39},\"visible-red-test-handoff\":{\"input\":890958,\"output\":9439,\"cacheRead\":6386176,\"cacheWrite\":0,\"turns\":35}},\"featureModels\":{\"retire-plan\":{\"openai-codex/gpt-5.6-terra\":{\"input\":243028,\"output\":842,\"cacheRead\":811520,\"cacheWrite\":0,\"turns\":9}},\"fresh-implementation-session\":{\"openai-codex/gpt-5.6-terra\":{\"input\":1086985,\"output\":12201,\"cacheRead\":6675968,\"cacheWrite\":0,\"turns\":32},\"openai-codex/gpt-5.6-sol\":{\"input\":1829390,\"output\":11330,\"cacheRead\":10031104,\"cacheWrite\":0,\"turns\":40}},\"active-plan-workspace\":{\"openai-codex/gpt-5.6-terra\":{\"input\":367195,\"output\":8243,\"cacheRead\":6820864,\"cacheWrite\":0,\"turns\":21},\"openai-codex/gpt-5.6-sol\":{\"input\":232073,\"output\":4848,\"cacheRead\":1564160,\"cacheWrite\":0,\"turns\":27}},\"settings-dashboard-modal\":{\"openai-codex/gpt-5.6-terra\":{\"input\":569517,\"output\":11997,\"cacheRead\":4862464,\"cacheWrite\":0,\"turns\":39}},\"visible-red-test-handoff\":{\"openai-codex/gpt-5.6-terra\":{\"input\":890958,\"output\":9439,\"cacheRead\":6386176,\"cacheWrite\":0,\"turns\":35}}}}"
+prices: "{}"
+timestamp: 2026-07-20T14:46:09.335Z
+---
+
+# Usage
+
+## hub
+
+| model | input | output | cache read | cache write | turns | cost |
+| --- | ---: | ---: | ---: | ---: | ---: | ---: |
+| openai-codex/gpt-5.6-terra | 243028 | 842 | 811520 | 0 | 9 | — |
+
+## plan
+
+| model | input | output | cache read | cache write | turns | cost |
+| --- | ---: | ---: | ---: | ---: | ---: | ---: |
+| openai-codex/gpt-5.6-sol | 118306 | 8575 | 3610112 | 0 | 21 | — |
+
+## implement
+
+| model | input | output | cache read | cache write | turns | cost |
+| --- | ---: | ---: | ---: | ---: | ---: | ---: |
+| openai-codex/gpt-5.6-terra | 2518150 | 40901 | 22285824 | 0 | 112 | — |
+| openai-codex/gpt-5.6-sol | 1332608 | 10554 | 7464960 | 0 | 42 | — |
+
+## review
+
+| model | input | output | cache read | cache write | turns | cost |
+| --- | ---: | ---: | ---: | ---: | ---: | ---: |
+| openai-codex/gpt-5.6-sol | 728855 | 5624 | 4130304 | 0 | 25 | — |
+| openai-codex/gpt-5.6-terra | 396505 | 979 | 2459648 | 0 | 15 | — |
+
+## Per feature
+
+| feature | input | output | cache read | cache write | turns | cost |
+| --- | ---: | ---: | ---: | ---: | ---: | ---: |
+| retire-plan | 243028 | 842 | 811520 | 0 | 9 | — |
+| fresh-implementation-session | 2916375 | 23531 | 16707072 | 0 | 72 | — |
+| active-plan-workspace | 599268 | 13091 | 8385024 | 0 | 48 | — |
+| settings-dashboard-modal | 569517 | 11997 | 4862464 | 0 | 39 | — |
+| visible-red-test-handoff | 890958 | 9439 | 6386176 | 0 | 35 | — |
+
+Total: 5337452 in / 67475 out / 40762368 cache-read / 0 cache-write over 224 turns. Cost unavailable: add every used model rate.
diff --git a/memory/features/archive/2026-07-20-focus-feature-execution-and-dashboard-ownership/visible-red-test-handoff.md b/memory/features/archive/2026-07-20-focus-feature-execution-and-dashboard-ownership/visible-red-test-handoff.md
new file mode 100644
index 0000000..d5372d0
--- /dev/null
+++ b/memory/features/archive/2026-07-20-focus-feature-execution-and-dashboard-ownership/visible-red-test-handoff.md
@@ -0,0 +1,50 @@
+---
+type: Feature
+title: Visible red-test handoff
+description: Show committed red tests as the implementation target in Work and carry their exact status and paths into the fresh implementer context.
+status: done
+size: medium
+depends_on: [active-plan-workspace]
+files: ["lib/gather.mjs", "lib/views/hub.mjs", "lib/pi-tools.mjs", "test/gather.test.mjs", "test/ui.test.mjs", "test/pi-tools.test.mjs"]
+memories: [pitfalls/client-js-template-literal-escaping, architecture/package-and-skill-layout, architecture/workflow-state-ownership, patterns/safe-browser-rendering, decisions/backlog-planning-and-feature-waves, decisions/iterator-dashboard-feature-workflow, decisions/manual-role-models-and-runtime-reset, decisions/memory-relevance-usage-and-dashboard-recovery]
+timestamp: "2026-07-20T14:45:56.689Z"
+tags: []
+commits:
+  - sha: 962693458648f7d084df7c41e1dedefcb5f8f509
+    kind: implement
+    date: 2026-07-20
+done: 2026-07-20
+reviewed: 2026-07-20
+---
+
+# Implementation notes
+
+Extend the server-gathered hub feature shape with recorded test paths/count where appropriate, preserving `tests_status` as the source of truth. In Work, make red an explicit expected checkpoint: badge/text should say tests are committed and intentionally failing, and the primary action should read as driving those tests green rather than generic implementation. After test creation, refresh/activate the relevant Work context so the new state is visible. Enrich the concise ambient/implementation kickoff state with the active feature's red status and test paths without duplicating the full contract or prior conversation. Ensure no UI or helper treats red as a failed implementation or silently changes tests; implementation remains responsible for the normal green gate.
+
+# Snippets
+
+```js
+return {
+  name: c.slug,
+  testsStatus: c.fm.tests_status || 'none',
+  // add the recorded test paths needed by Work and the handoff summary
+};
+```
+
+```js
+test.textContent = c.status==='pending' ? 'Test (red)' : 'Test';
+impl.textContent = 'Implement';
+```
+
+# Depends on
+
+* [Active plan workspace](/features/active-plan-workspace.md)
+
+# Blast radius
+
+Feature cards, footer/ambient context size, gather payload compatibility, and the red/green contract used by manual and automatic implementation.
+
+# Review
+
+## 2026-07-20
+* **Approved** — Accepted through the visible red-test handoff review.
diff --git a/memory/features/archive/2026-07-20-safe-role-model-restoration/index.md b/memory/features/archive/2026-07-20-safe-role-model-restoration/index.md
new file mode 100644
index 0000000..bc527db
--- /dev/null
+++ b/memory/features/archive/2026-07-20-safe-role-model-restoration/index.md
@@ -0,0 +1,3 @@
+# Features
+
+* [Safe role-model handoff](safe-role-model-handoff.md) - ✅ done · medium · Keep the active managed model usable when a tester or implementer role override is unavailable, so red-test completion can hand off to implementation without an OpenAI credential error.
diff --git a/memory/features/archive/2026-07-20-safe-role-model-restoration/plan.md b/memory/features/archive/2026-07-20-safe-role-model-restoration/plan.md
new file mode 100644
index 0000000..889cbfc
--- /dev/null
+++ b/memory/features/archive/2026-07-20-safe-role-model-restoration/plan.md
@@ -0,0 +1,36 @@
+---
+type: Plan
+title: Fix post-test role model credential corruption
+description: Prevent red-test completion and role-model restoration from leaving the next Iterator turn on an unusable OpenAI credential path.
+status: approved
+branch: iterator/fix-post-test-role-model-credential-corruption
+worktree: /Volumes/Extern/Projects/iterator-iterator-fix-post-test-role-model-credential-corruption
+created: 2026-07-20
+timestamp: 2026-07-20T13:10:00.129Z
+---
+
+# Goal
+
+Make the red-test → implementation handoff reliable: after `/iterator-test` writes and commits intentionally failing tests, the following `/iterator-implement` turn must use the intended active or explicitly configured role model without producing a `proxy-managed` OpenAI 401 or leaking model state between turns.
+
+# Architecture
+
+- Build on `architecture/package-and-skill-layout`: isolate the extension's manual-role lifecycle (`input` → `before_agent_start` → `agent_end`) and reproduce the tester-to-implementer sequence with direct extension-hook tests, including managed/proxied active models and failed switches.
+- Keep role resolution in `lib/pi-tools.mjs` and session orchestration in `extensions/iterator.js`; make role state turn-scoped, consume stale pending roles, and only restore a prior model when an explicit switch actually succeeded.
+- Treat `active` as a strict no-model-switch path. Preserve thinking-level overrides independently, and never synthesize, export, or rewrite provider API-key environment variables.
+- Verify both manual red-test handoff and automatic/wave role transitions, then run `npm run sync` for any canonical `lib/` changes so shipped skill copies remain identical.
+
+# Dependencies
+
+(none)
+
+# Key decisions
+
+- Follow `decisions/manual-role-models-and-runtime-reset`: retain configured per-role model selection and restoration, but harden its lifecycle rather than disabling role models.
+- A default or explicit `active` role model must not call `pi.setModel`; an unknown, unavailable, or failed explicit override must stay on the usable current model and must not arm a later restoration.
+- Add hook-level regression coverage for tester → implementer and restoration failure paths, closing the direct extension-lifecycle test gap recorded by `decisions/manual-role-models-and-runtime-reset`.
+- Preserve the red/green feature contract: red tests remain committed and visible to implementation; deferred formatter output is treated as context, not as a trigger for provider selection.
+
+# Features
+
+* [Safe role-model handoff](/features/safe-role-model-handoff.md) - Keep the active managed model usable when a tester or implementer role override is unavailable, so red-test completion can hand off to implementation without an OpenAI credential error.
diff --git a/memory/features/archive/2026-07-20-safe-role-model-restoration/safe-role-model-handoff.md b/memory/features/archive/2026-07-20-safe-role-model-restoration/safe-role-model-handoff.md
new file mode 100644
index 0000000..9aca296
--- /dev/null
+++ b/memory/features/archive/2026-07-20-safe-role-model-restoration/safe-role-model-handoff.md
@@ -0,0 +1,48 @@
+---
+type: Feature
+title: Safe role-model handoff
+description: Keep the active managed model usable when a tester or implementer role override is unavailable, so red-test completion can hand off to implementation without an OpenAI credential error.
+status: done
+size: medium
+depends_on: []
+files: ["extensions/iterator.js", "test/extension-model-lifecycle.test.mjs"]
+memories: [architecture/package-and-skill-layout, decisions/backlog-planning-and-feature-waves, decisions/iterator-dashboard-feature-workflow, decisions/manual-role-models-and-runtime-reset, decisions/memory-relevance-usage-and-dashboard-recovery, decisions/parallel-feature-waves-and-consolidated-review, decisions/polish-dashboard-and-multi-agent-workflows, decisions/powerline-shows-sandbox-ui-port]
+timestamp: "2026-07-20T13:33:09.759Z"
+tags: []
+commits:
+  - sha: 9116540c8f0bdac6dd7bab5041c9f3245981daf8
+    kind: implement
+    date: 2026-07-20
+done: 2026-07-20
+---
+
+# Implementation notes
+
+Reproduce the exact manual lifecycle through mocked Pi hooks: `/iterator-test` input, `before_agent_start`, failed or unavailable `pi.setModel`, `agent_end`, then `/iterator-implement`. In `extensions/iterator.js`, consume the pending role per turn and distinguish an attempted switch from a successful switch. Do not arm `preAutoModel` or invoke restoration unless `pi.setModel` returned success; keep thinking-level handling independent. Cover `active` settings as a strict no-`setModel` path, a failed explicit tester override followed by implementation, a successful override restoring once, and stale-role isolation. Preserve the same behavior for automatic/wave roles and do not alter environment variables or red-test artifacts.
+
+# Snippets
+
+```js
+const applyRole = async (role, settings) => {
+  const spec = roleModelSpec(settings, role);
+  // Current bug seam: preAutoModel is armed before setModel reports success.
+  const ok = await pi.setModel(model);
+};
+```
+
+```js
+pi.on("before_agent_start", async (_event, ctx) => {
+  if (pendingRole && state?.mode !== "auto" && !featureWave) {
+    await applyRole(pendingRole, settings);
+    manualRoleActive = true;
+  }
+});
+
+pi.on("agent_end", async (_event, ctx) => {
+  if (manualRoleActive) await restoreModel();
+});
+```
+
+# Blast radius
+
+Manual Iterator role commands plus auto-mode and ready-wave model transitions; incorrect restoration can affect the provider credentials used by every subsequent agent turn.
diff --git a/memory/features/archive/2026-07-20-safe-role-model-restoration/usage.md b/memory/features/archive/2026-07-20-safe-role-model-restoration/usage.md
new file mode 100644
index 0000000..78582d3
--- /dev/null
+++ b/memory/features/archive/2026-07-20-safe-role-model-restoration/usage.md
@@ -0,0 +1,43 @@
+---
+type: Usage
+title: Token usage
+description: Per-step model/token ledger and optional project-owned pricing for the active plan — written only by the usage op.
+totals: "{\"steps\":{\"hub\":{\"openai-codex/gpt-5.6-terra\":{\"input\":261034,\"output\":1116,\"cacheRead\":1795072,\"cacheWrite\":0,\"turns\":8}},\"plan\":{\"openai-codex/gpt-5.6-sol\":{\"input\":63084,\"output\":4999,\"cacheRead\":612864,\"cacheWrite\":0,\"turns\":20}},\"implement\":{\"openai-codex/gpt-5.6-terra\":{\"input\":117926,\"output\":9423,\"cacheRead\":2369024,\"cacheWrite\":0,\"turns\":31}},\"review\":{\"openai-codex/gpt-5.6-sol\":{\"input\":161613,\"output\":576,\"cacheRead\":722944,\"cacheWrite\":0,\"turns\":8}}},\"features\":{\"retire-plan\":{\"input\":261034,\"output\":1116,\"cacheRead\":1795072,\"cacheWrite\":0,\"turns\":8},\"safe-role-model-handoff\":{\"input\":279539,\"output\":9999,\"cacheRead\":3091968,\"cacheWrite\":0,\"turns\":39}},\"featureModels\":{\"safe-role-model-handoff\":{\"openai-codex/gpt-5.6-terra\":{\"input\":117926,\"output\":9423,\"cacheRead\":2369024,\"cacheWrite\":0,\"turns\":31},\"openai-codex/gpt-5.6-sol\":{\"input\":161613,\"output\":576,\"cacheRead\":722944,\"cacheWrite\":0,\"turns\":8}}}}"
+prices: "{}"
+timestamp: 2026-07-20T13:35:07.863Z
+---
+
+# Usage
+
+## hub
+
+| model | input | output | cache read | cache write | turns | cost |
+| --- | ---: | ---: | ---: | ---: | ---: | ---: |
+| openai-codex/gpt-5.6-terra | 261034 | 1116 | 1795072 | 0 | 8 | — |
+
+## plan
+
+| model | input | output | cache read | cache write | turns | cost |
+| --- | ---: | ---: | ---: | ---: | ---: | ---: |
+| openai-codex/gpt-5.6-sol | 63084 | 4999 | 612864 | 0 | 20 | — |
+
+## implement
+
+| model | input | output | cache read | cache write | turns | cost |
+| --- | ---: | ---: | ---: | ---: | ---: | ---: |
+| openai-codex/gpt-5.6-terra | 117926 | 9423 | 2369024 | 0 | 31 | — |
+
+## review
+
+| model | input | output | cache read | cache write | turns | cost |
+| --- | ---: | ---: | ---: | ---: | ---: | ---: |
+| openai-codex/gpt-5.6-sol | 161613 | 576 | 722944 | 0 | 8 | — |
+
+## Per feature
+
+| feature | input | output | cache read | cache write | turns | cost |
+| --- | ---: | ---: | ---: | ---: | ---: | ---: |
+| retire-plan | 261034 | 1116 | 1795072 | 0 | 8 | — |
+| safe-role-model-handoff | 279539 | 9999 | 3091968 | 0 | 39 | — |
+
+Total: 603657 in / 16114 out / 5499904 cache-read / 0 cache-write over 67 turns. Cost unavailable: add every used model rate.
diff --git a/memory/features/archive/2026-07-20-streamline-backlog-planning-and-knowledge-actions/backlog-file-mentions.md b/memory/features/archive/2026-07-20-streamline-backlog-planning-and-knowledge-actions/backlog-file-mentions.md
new file mode 100644
index 0000000..38c836b
--- /dev/null
+++ b/memory/features/archive/2026-07-20-streamline-backlog-planning-and-knowledge-actions/backlog-file-mentions.md
@@ -0,0 +1,37 @@
+---
+type: Feature
+title: Backlog file mentions
+description: "Add Planning-style @ file suggestions to backlog idea details so candidates can anchor repository context."
+status: done
+size: small
+depends_on: []
+files: ["lib/views/planning.mjs", "test/ui.test.mjs", "test/client-js-parse.test.mjs"]
+memories: [pitfalls/client-js-template-literal-escaping, architecture/workflow-state-ownership, decisions/backlog-planning-and-feature-waves, decisions/consume-accepted-backlog-ideas, decisions/focus-feature-execution-and-dashboard-ownership, decisions/iterator-dashboard-feature-workflow, decisions/memory-relevance-usage-and-dashboard-recovery, decisions/parallel-feature-waves-and-consolidated-review]
+timestamp: "2026-07-20T14:56:05.501Z"
+tags: []
+commits:
+  - sha: c4bac704a6cd574de68ff95a67f3cee2f894c947
+    kind: implement
+    date: 2026-07-20
+done: 2026-07-20
+reviewed: 2026-07-20
+---
+
+# Implementation notes
+
+Reuse the safe Planning @-suggestion behavior in the backlog details control. Keep stored candidate text plain and preserve inline-script escaping and mobile controls.
+
+# Snippets
+
+```js
+wireAtMenu(goal, atMenu);
+```
+
+# Blast radius
+
+Backlog candidate entry and client-script parse safety.
+
+# Review
+
+## 2026-07-20
+* **Approved** _(agent review: openai-codex/gpt-5.6-sol)_ — Backlog details now reuse the existing escaped, keyboard-accessible @ file menu, retain plain-text persistence, and keep responsive and inline-script parse coverage green.
diff --git a/memory/features/archive/2026-07-20-streamline-backlog-planning-and-knowledge-actions/backlog-filter-and-bulk-select.md b/memory/features/archive/2026-07-20-streamline-backlog-planning-and-knowledge-actions/backlog-filter-and-bulk-select.md
new file mode 100644
index 0000000..ecc2748
--- /dev/null
+++ b/memory/features/archive/2026-07-20-streamline-backlog-planning-and-knowledge-actions/backlog-filter-and-bulk-select.md
@@ -0,0 +1,46 @@
+---
+type: Feature
+title: Backlog filtering and bulk selection
+description: Filter backlog candidates by type and select or deselect the visible set for planning.
+status: done
+size: medium
+depends_on: [backlog-save-during-work]
+files: ["lib/views/planning.mjs", "lib/write.mjs", "test/ui.test.mjs", "test/write.test.mjs"]
+memories: [architecture/workflow-state-ownership, decisions/backlog-planning-and-feature-waves, decisions/consume-accepted-backlog-ideas, decisions/focus-feature-execution-and-dashboard-ownership, decisions/iterator-dashboard-feature-workflow, decisions/manual-role-models-and-runtime-reset, decisions/memory-relevance-usage-and-dashboard-recovery, decisions/parallel-feature-waves-and-consolidated-review]
+timestamp: "2026-07-20T15:08:53.196Z"
+tags: []
+tests_status: green
+commits:
+  - sha: f169ba883dc43251afd128dc14a62c80928b1829
+    kind: implement
+    date: 2026-07-20
+  - sha: 0e543d11713adc5da35ed32a95b095472a2048fe
+    kind: implement
+    date: 2026-07-20
+reviewed: 2026-07-20
+done: 2026-07-20
+---
+
+# Implementation notes
+
+Render client-side type filters and visible-set select/deselect controls. Persist each selection through existing deterministic backlog writes; never consume candidates until plan approval.
+
+# Snippets
+
+```js
+backlogAction({ action: 'select', id: item.id, selected: !item.selected }, button)
+```
+
+# Depends on
+
+* [Reliable backlog saves](/features/backlog-save-during-work.md)
+
+# Blast radius
+
+Backlog selection semantics, plan candidate consumption, and responsive Planning controls.
+
+# Review
+
+## 2026-07-20
+* **Approved** _(agent review: openai-codex/gpt-5.6-sol)_ — Approved: filtering targets only visible candidates, bulk selection now persists all IDs in one validated deterministic write, and UI, writer, extension, and session coverage verifies the handoff without weakening plan-approval consumption.
+* **Needs changes** _(agent review: openai-codex/gpt-5.6-sol)_ — Bulk select is not durably serialized: each per-item post receives HTTP 200 before the async saveBacklog write finishes, so the loop launches concurrent read-modify-write operations against memory/backlog/index.md; updates can overwrite one another and each save also triggers a dashboard refresh. Send the visible IDs through one acknowledged deterministic bulk operation (or otherwise serialize server-side writes and refresh once), and add a regression test through the session/extension boundary proving a multi-item visible selection persists every item.
diff --git a/memory/features/archive/2026-07-20-streamline-backlog-planning-and-knowledge-actions/backlog-save-during-work.md b/memory/features/archive/2026-07-20-streamline-backlog-planning-and-knowledge-actions/backlog-save-during-work.md
new file mode 100644
index 0000000..10b546c
--- /dev/null
+++ b/memory/features/archive/2026-07-20-streamline-backlog-planning-and-knowledge-actions/backlog-save-during-work.md
@@ -0,0 +1,37 @@
+---
+type: Feature
+title: Reliable backlog saves
+description: Keep deterministic backlog saves available during active work without clearing the owned Work overlay.
+status: done
+size: medium
+depends_on: []
+files: ["lib/session-server.mjs", "extensions/iterator.js", "test/session-server.test.mjs"]
+memories: [pitfalls/cancel-now-after-grace-timer, architecture/browser-server-contract, architecture/package-and-skill-layout, decisions/backlog-planning-and-feature-waves, decisions/focus-feature-execution-and-dashboard-ownership, decisions/iterator-dashboard-feature-workflow, decisions/manual-role-models-and-runtime-reset, decisions/memory-relevance-usage-and-dashboard-recovery]
+timestamp: "2026-07-20T14:59:18.522Z"
+tags: []
+commits:
+  - sha: e5f2027ab7e5d4c51af978ec1c7601633122ca4c
+    kind: implement
+    date: 2026-07-20
+done: 2026-07-20
+reviewed: 2026-07-20
+---
+
+# Implementation notes
+
+Allow only the existing backlog write protocol while the agent is working, route it through the deterministic writer, and refresh stored dashboard data without ending or replacing the active work claim. Do not broaden the exception to model-starting actions.
+
+# Snippets
+
+```js
+const backlogWrite = parsed.type === 'backlog';
+```
+
+# Blast radius
+
+Single-model-flow guard, Work overlay ownership, and backlog persistence.
+
+# Review
+
+## 2026-07-20
+* **Approved** _(agent review: openai-codex/gpt-5.6-sol)_ — Backlog CRUD remains the sole filesystem-write exception during active work, refreshed Planning state is stored immediately, and the owned Work guard plus model-flow blocking remain intact.
diff --git a/memory/features/archive/2026-07-20-streamline-backlog-planning-and-knowledge-actions/full-plan-fast-track.md b/memory/features/archive/2026-07-20-streamline-backlog-planning-and-knowledge-actions/full-plan-fast-track.md
new file mode 100644
index 0000000..0fd732c
--- /dev/null
+++ b/memory/features/archive/2026-07-20-streamline-backlog-planning-and-knowledge-actions/full-plan-fast-track.md
@@ -0,0 +1,37 @@
+---
+type: Feature
+title: Full-plan fast track
+description: Turn a structured plan supplied in Planning into an approved feature-ready draft without redundant planning prompts.
+status: done
+size: medium
+depends_on: []
+files: ["lib/views/planning.mjs", "lib/views/plan.mjs", "extensions/iterator.js", "test/ui.test.mjs", "test/session-server.test.mjs"]
+memories: [pitfalls/cancel-now-after-grace-timer, architecture/package-and-skill-layout, architecture/workflow-state-ownership, decisions/backlog-planning-and-feature-waves, decisions/consume-accepted-backlog-ideas, decisions/focus-feature-execution-and-dashboard-ownership, decisions/iterator-dashboard-feature-workflow, decisions/manual-role-models-and-runtime-reset]
+timestamp: "2026-07-20T15:15:01.641Z"
+tags: []
+commits:
+  - sha: 48f44eb603a2c55b1585f841344c552a531d58ef
+    kind: implement
+    date: 2026-07-20
+done: 2026-07-20
+reviewed: 2026-07-20
+---
+
+# Implementation notes
+
+Detect a structured plan submission at the Planning entry point, preserve the browser plan-review and deterministic plan writer, then hand the approved result directly to the feature breakdown flow. Keep ordinary goal-only planning unchanged and do not bypass candidate consumption or approval.
+
+# Snippets
+
+```js
+post({ type:'action', action:'plan', feature:null, prompt: goal.value.trim() || null }, 'Starting /iterator-plan')
+```
+
+# Blast radius
+
+Planning entry behavior, plan approval continuation, and fresh-session dashboard context.
+
+# Review
+
+## 2026-07-20
+* **Approved** _(agent review: openai-codex/gpt-5.6-sol)_ — Approved: canonical structured plans bypass only redundant planner discovery, remain editable in browser review, pass through the deterministic plan writer and approval-bound backlog consumption, and dispatch feature breakdown only after a successful approved write.
diff --git a/memory/features/archive/2026-07-20-streamline-backlog-planning-and-knowledge-actions/index.md b/memory/features/archive/2026-07-20-streamline-backlog-planning-and-knowledge-actions/index.md
new file mode 100644
index 0000000..d3d1d61
--- /dev/null
+++ b/memory/features/archive/2026-07-20-streamline-backlog-planning-and-knowledge-actions/index.md
@@ -0,0 +1,8 @@
+# Features
+
+* [Backlog file mentions](backlog-file-mentions.md) - ✅ done · small · Add Planning-style @ file suggestions to backlog idea details so candidates can anchor repository context.
+* [Reliable backlog saves](backlog-save-during-work.md) - ✅ done · medium · Keep deterministic backlog saves available during active work without clearing the owned Work overlay.
+* [Backlog filtering and bulk selection](backlog-filter-and-bulk-select.md) - ✅ done · 🟢 tests green · medium · depends: backlog-save-during-work · Filter backlog candidates by type and select or deselect the visible set for planning.
+* [Full-plan fast track](full-plan-fast-track.md) - ✅ done · medium · Turn a structured plan supplied in Planning into an approved feature-ready draft without redundant planning prompts.
+* [Focused memory review changes](memory-review-change-focus.md) - ✅ done · medium · Highlight changed sections in memory proposals so reviewers can see exactly what a create or update will alter.
+* [Work conflict memory resolution](work-conflict-memory-resolution.md) - ✅ done · medium · depends: memory-review-change-focus · Let a Work feature conflict initiate a reviewed decision-update request and anchored follow-up check.
diff --git a/memory/features/archive/2026-07-20-streamline-backlog-planning-and-knowledge-actions/memory-review-change-focus.md b/memory/features/archive/2026-07-20-streamline-backlog-planning-and-knowledge-actions/memory-review-change-focus.md
new file mode 100644
index 0000000..1d99ecf
--- /dev/null
+++ b/memory/features/archive/2026-07-20-streamline-backlog-planning-and-knowledge-actions/memory-review-change-focus.md
@@ -0,0 +1,41 @@
+---
+type: Feature
+title: Focused memory review changes
+description: Highlight changed sections in memory proposals so reviewers can see exactly what a create or update will alter.
+status: done
+size: medium
+depends_on: []
+files: ["lib/views/memory-review.mjs", "lib/ui.mjs", "test/ui.test.mjs", "test/client-js-parse.test.mjs"]
+memories: [pitfalls/client-js-template-literal-escaping, architecture/browser-server-contract, patterns/agent-reviewed-memory-writes, patterns/safe-browser-rendering, decisions/iterator-dashboard-feature-workflow, decisions/parallel-feature-waves-and-consolidated-review, decisions/polish-dashboard-and-multi-agent-workflows, decisions/review-navigation-and-work-context]
+timestamp: "2026-07-20T15:22:20.347Z"
+tags: []
+commits:
+  - sha: dff652ec86deb8aaba625755a71e75ac4bb2f563
+    kind: implement
+    date: 2026-07-20
+  - sha: 9213ff8661b525aa94d53cbaaf403c01e21f4c5f
+    kind: implement
+    date: 2026-07-20
+reviewed: 2026-07-20
+done: 2026-07-20
+---
+
+# Implementation notes
+
+Compare proposed and existing concept bodies in the browser using safe escaped rendering. Make changed blocks visually distinct while retaining the existing review verdict and writer protocol.
+
+# Snippets
+
+```js
+hydrateMemoryCards(payload, ctx.cwd);
+```
+
+# Blast radius
+
+Knowledge review readability, embedded data safety, and mobile diff layout.
+
+# Review
+
+## 2026-07-20
+* **Approved** _(agent review: openai-codex/gpt-5.6-sol)_ — Approved: create and update proposals clearly distinguish added, modified, removed, and unchanged markdown sections; fenced content and embedded data stay safe, verdict/writer protocols are unchanged, and the executed href regression verifies quote safety without corrupting query parameters.
+* **Needs changes** _(agent review: openai-codex/gpt-5.6-sol)_ — The new href hardening double-escapes ampersands in valid links: inline() first runs escc(s), so the captured URL already contains &amp;, then attr(u) runs escc again and emits &amp;amp;, changing query-string URLs in the browser. Escape only quote characters at the attribute step (or restructure link parsing to escape exactly once), and add an executed regression proving an allowed URL with both '&' and quotes is safe without changing its query parameters.
diff --git a/memory/features/archive/2026-07-20-streamline-backlog-planning-and-knowledge-actions/plan.md b/memory/features/archive/2026-07-20-streamline-backlog-planning-and-knowledge-actions/plan.md
new file mode 100644
index 0000000..d7ec3cd
--- /dev/null
+++ b/memory/features/archive/2026-07-20-streamline-backlog-planning-and-knowledge-actions/plan.md
@@ -0,0 +1,55 @@
+---
+type: Plan
+title: Streamline backlog planning and knowledge actions
+description: Make backlog-to-plan creation, knowledge review, and active-work conflict handling faster and more visible.
+status: approved
+branch: iterator/safe-role-model-handoff
+created: 2026-07-20
+timestamp: "2026-07-20T15:33:07.184Z"
+plan_reviewed: 2026-07-20
+---
+
+# Goal
+
+Turn the selected backlog ideas into a faster, safer planning and knowledge workflow: accept a supplied full plan as a feature-ready plan draft, make backlog capture and filtering reliable, surface focused memory changes, and provide a reviewed path from Work conflict warnings to decision updates.
+
+# Architecture
+
+- Extend the existing Planning/backlog views and server-derived gather payloads rather than creating parallel workflow state; keep candidate filtering, bulk selection, and `@` file suggestions in the client view while deterministic backlog writes remain in `lib/write.mjs`.
+- Treat a structured user-supplied plan as a prefilled plan draft that still receives the existing browser approval, then continue directly to `/iterator-feature`; do not bypass the deterministic plan writer, feature slicing, dependency gates, or candidate-consumption rules.
+- Reuse the persistent shell and session-server contract so filesystem-only backlog saves remain available during active work without clearing its Work overlay or starting a second model flow.
+- Extend the reviewed memory UI and knowledge/write paths to expose changed sections in memory proposals and offer a Work conflict action that creates a reviewable decision-update request; never edit a decision directly from a feature card.
+- Keep lifecycle/readiness and conflict status server-derived through `lib/status.mjs` and gather payloads. Preserve safe JSON embedding, client-script parse coverage, responsive controls, and root-to-skill synchronization for canonical `lib/` changes.
+
+# Dependencies
+
+(none)
+
+# Key decisions
+
+- A full plan input skips only redundant planning questions: browser review and deterministic plan approval remain mandatory before features are created.
+- Backlog save, filter, and bulk-selection controls remain deterministic filesystem actions and may run while an agent works, but they must not unblock or start another model flow.
+- Selected backlog candidates are consumed only after approved plan creation, per `decisions/iterator-dashboard-feature-workflow`; cancelled or feedback plan rounds retain them.
+- Memory-conflict acceptance is a reviewed knowledge workflow: it proposes a decision update and an anchored recheck, rather than silently changing memory or marking conflicts resolved.
+- Memory-review highlighting distinguishes changed sections from unchanged text without exposing unsafe raw HTML or changing the review acceptance gate.
+- Follow `memory/design.md`: compact responsive controls, semantic status colors, and no view-local lifecycle inference.
+
+# Features
+
+* [Backlog file mentions](/features/backlog-file-mentions.md) - Add Planning-style @ file suggestions to backlog idea details so candidates can anchor repository context.
+* [Reliable backlog saves](/features/backlog-save-during-work.md) - Keep deterministic backlog saves available during active work without clearing the owned Work overlay.
+* [Backlog filtering and bulk selection](/features/backlog-filter-and-bulk-select.md) - Filter backlog candidates by type and select or deselect the visible set for planning.
+* [Full-plan fast track](/features/full-plan-fast-track.md) - Turn a structured plan supplied in Planning into an approved feature-ready draft without redundant planning prompts.
+* [Focused memory review changes](/features/memory-review-change-focus.md) - Highlight changed sections in memory proposals so reviewers can see exactly what a create or update will alter.
+* [Work conflict memory resolution](/features/work-conflict-memory-resolution.md) - Let a Work feature conflict initiate a reviewed decision-update request and anchored follow-up check.
+
+# Plan review
+
+## 2026-07-20 _(agent review: openai-codex/gpt-5.6-sol)_
+
+## Clean bill
+
+- **Goal coverage:** All stated workflow goals are represented in the eight feature commits: structured plans enter direct editable approval before deterministic writing and feature slicing; backlog details support file mentions; backlog saves remain available under the owned Work guard; type filters and atomic visible-set selection are present; memory proposals focus added, modified, removed, and unchanged sections; and Work conflict cards route decision updates through reviewed knowledge handling with anchored follow-up guidance.
+- **Key decisions:** Approval-bound backlog consumption remains in the plan writer; ordinary goals retain the planner path; bulk selection uses one validated filesystem write; memory changes still require explicit verdicts; conflict handling does not directly edit decisions or clear unrelated conflict flags; and safe embedding/markdown-link regressions cover the new browser rendering.
+- **Architecture:** Changes extend the existing Planning, Work, gather, writer, persistent-session, and memory-review seams. Server-derived conflict metadata supplies the Work view, while canonical root libraries are synchronized to shipped skill copies.
+- **Scope and loose ends:** No unexplained production behavior, new dependency, introduced TODO marker, or unaccepted feature remains in the committed plan scope. Rework commits address both recorded feature-review findings. Existing unrelated working-tree leftovers were excluded from feature commits and are not part of this plan review.
diff --git a/memory/features/archive/2026-07-20-streamline-backlog-planning-and-knowledge-actions/usage.md b/memory/features/archive/2026-07-20-streamline-backlog-planning-and-knowledge-actions/usage.md
new file mode 100644
index 0000000..d59cdda
--- /dev/null
+++ b/memory/features/archive/2026-07-20-streamline-backlog-planning-and-knowledge-actions/usage.md
@@ -0,0 +1,49 @@
+---
+type: Usage
+title: Token usage
+description: Per-step model/token ledger and optional project-owned pricing for the active plan — written only by the usage op.
+totals: "{\"steps\":{\"hub\":{\"openai-codex/gpt-5.6-terra\":{\"input\":39930,\"output\":2073,\"cacheRead\":2132480,\"cacheWrite\":0,\"turns\":8}},\"plan\":{\"openai-codex/gpt-5.6-terra\":{\"input\":21413,\"output\":4965,\"cacheRead\":3170816,\"cacheWrite\":0,\"turns\":11}},\"implement\":{\"openai-codex/gpt-5.6-terra\":{\"input\":320231,\"output\":2436,\"cacheRead\":4639744,\"cacheWrite\":0,\"turns\":16},\"openai-codex/gpt-5.6-sol\":{\"input\":1550065,\"output\":45118,\"cacheRead\":38785024,\"cacheWrite\":0,\"turns\":197}},\"review\":{\"openai-codex/gpt-5.6-sol\":{\"input\":1758022,\"output\":7246,\"cacheRead\":10731520,\"cacheWrite\":0,\"turns\":59}}},\"features\":{\"retire-plan\":{\"input\":39930,\"output\":2073,\"cacheRead\":2132480,\"cacheWrite\":0,\"turns\":8},\"backlog-file-mentions\":{\"input\":631890,\"output\":2836,\"cacheRead\":5563904,\"cacheWrite\":0,\"turns\":20},\"backlog-save-during-work\":{\"input\":717690,\"output\":4017,\"cacheRead\":5634048,\"cacheWrite\":0,\"turns\":20},\"backlog-filter-and-bulk-select\":{\"input\":514198,\"output\":13256,\"cacheRead\":8431104,\"cacheWrite\":0,\"turns\":73},\"full-plan-fast-track\":{\"input\":333812,\"output\":11344,\"cacheRead\":6500352,\"cacheWrite\":0,\"turns\":44},\"memory-review-change-focus\":{\"input\":625745,\"output\":9383,\"cacheRead\":10724864,\"cacheWrite\":0,\"turns\":53},\"work-conflict-memory-resolution\":{\"input\":753984,\"output\":12866,\"cacheRead\":14944256,\"cacheWrite\":0,\"turns\":55}},\"featureModels\":{\"retire-plan\":{\"openai-codex/gpt-5.6-terra\":{\"input\":39930,\"output\":2073,\"cacheRead\":2132480,\"cacheWrite\":0,\"turns\":8}},\"backlog-file-mentions\":{\"openai-codex/gpt-5.6-terra\":{\"input\":320231,\"output\":2436,\"cacheRead\":4639744,\"cacheWrite\":0,\"turns\":16},\"openai-codex/gpt-5.6-sol\":{\"input\":311659,\"output\":400,\"cacheRead\":924160,\"cacheWrite\":0,\"turns\":4}},\"backlog-save-during-work\":{\"openai-codex/gpt-5.6-sol\":{\"input\":717690,\"output\":4017,\"cacheRead\":5634048,\"cacheWrite\":0,\"turns\":20}},\"backlog-filter-and-bulk-select\":{\"openai-codex/gpt-5.6-sol\":{\"input\":514198,\"output\":13256,\"cacheRead\":8431104,\"cacheWrite\":0,\"turns\":73}},\"full-plan-fast-track\":{\"openai-codex/gpt-5.6-sol\":{\"input\":333812,\"output\":11344,\"cacheRead\":6500352,\"cacheWrite\":0,\"turns\":44}},\"memory-review-change-focus\":{\"openai-codex/gpt-5.6-sol\":{\"input\":625745,\"output\":9383,\"cacheRead\":10724864,\"cacheWrite\":0,\"turns\":53}},\"work-conflict-memory-resolution\":{\"openai-codex/gpt-5.6-sol\":{\"input\":753984,\"output\":12866,\"cacheRead\":14944256,\"cacheWrite\":0,\"turns\":55}}}}"
+prices: "{}"
+timestamp: 2026-07-20T15:33:24.438Z
+---
+
+# Usage
+
+## hub
+
+| model | input | output | cache read | cache write | turns | cost |
+| --- | ---: | ---: | ---: | ---: | ---: | ---: |
+| openai-codex/gpt-5.6-terra | 39930 | 2073 | 2132480 | 0 | 8 | — |
+
+## plan
+
+| model | input | output | cache read | cache write | turns | cost |
+| --- | ---: | ---: | ---: | ---: | ---: | ---: |
+| openai-codex/gpt-5.6-terra | 21413 | 4965 | 3170816 | 0 | 11 | — |
+
+## implement
+
+| model | input | output | cache read | cache write | turns | cost |
+| --- | ---: | ---: | ---: | ---: | ---: | ---: |
+| openai-codex/gpt-5.6-terra | 320231 | 2436 | 4639744 | 0 | 16 | — |
+| openai-codex/gpt-5.6-sol | 1550065 | 45118 | 38785024 | 0 | 197 | — |
+
+## review
+
+| model | input | output | cache read | cache write | turns | cost |
+| --- | ---: | ---: | ---: | ---: | ---: | ---: |
+| openai-codex/gpt-5.6-sol | 1758022 | 7246 | 10731520 | 0 | 59 | — |
+
+## Per feature
+
+| feature | input | output | cache read | cache write | turns | cost |
+| --- | ---: | ---: | ---: | ---: | ---: | ---: |
+| retire-plan | 39930 | 2073 | 2132480 | 0 | 8 | — |
+| backlog-file-mentions | 631890 | 2836 | 5563904 | 0 | 20 | — |
+| backlog-save-during-work | 717690 | 4017 | 5634048 | 0 | 20 | — |
+| backlog-filter-and-bulk-select | 514198 | 13256 | 8431104 | 0 | 73 | — |
+| full-plan-fast-track | 333812 | 11344 | 6500352 | 0 | 44 | — |
+| memory-review-change-focus | 625745 | 9383 | 10724864 | 0 | 53 | — |
+| work-conflict-memory-resolution | 753984 | 12866 | 14944256 | 0 | 55 | — |
+
+Total: 3689661 in / 61838 out / 59459584 cache-read / 0 cache-write over 291 turns. Cost unavailable: add every used model rate.
diff --git a/memory/features/archive/2026-07-20-streamline-backlog-planning-and-knowledge-actions/work-conflict-memory-resolution.md b/memory/features/archive/2026-07-20-streamline-backlog-planning-and-knowledge-actions/work-conflict-memory-resolution.md
new file mode 100644
index 0000000..d04c573
--- /dev/null
+++ b/memory/features/archive/2026-07-20-streamline-backlog-planning-and-knowledge-actions/work-conflict-memory-resolution.md
@@ -0,0 +1,41 @@
+---
+type: Feature
+title: Work conflict memory resolution
+description: Let a Work feature conflict initiate a reviewed decision-update request and anchored follow-up check.
+status: done
+size: medium
+depends_on: [memory-review-change-focus]
+files: ["lib/views/hub.mjs", "lib/views/memory-review.mjs", "extensions/iterator.js", "lib/gather.mjs", "test/ui.test.mjs", "test/gather.test.mjs"]
+memories: [pitfalls/client-js-template-literal-escaping, architecture/package-and-skill-layout, architecture/workflow-state-ownership, patterns/agent-reviewed-memory-writes, patterns/safe-browser-rendering, decisions/backlog-planning-and-feature-waves, decisions/focus-feature-execution-and-dashboard-ownership, decisions/iterator-dashboard-feature-workflow]
+timestamp: "2026-07-20T15:32:13.604Z"
+tags: []
+commits:
+  - sha: 0285ceb2302d7c7e66ffc76465fd6aac453126e8
+    kind: implement
+    date: 2026-07-20
+done: 2026-07-20
+reviewed: 2026-07-20
+---
+
+# Implementation notes
+
+Add a conflict-card action that opens the existing knowledge review path with the conflicting decision and feature context. Require an explicit memory verdict before updating durable knowledge; refresh gather-derived conflict flags afterward without auto-resolving unrelated features.
+
+# Snippets
+
+```js
+(c.conflicts?'<span class="chip cr" title="This feature contradicts a project decision">...':'')
+```
+
+# Depends on
+
+* [Focused memory review changes](/features/memory-review-change-focus.md)
+
+# Blast radius
+
+Work conflict cards, reviewed knowledge writes, and feature conflict gathering.
+
+# Review
+
+## 2026-07-20
+* **Approved** _(agent review: openai-codex/gpt-5.6-sol)_ — Approved: Work surfaces exact decision conflicts with anchored context, routes updates through the existing explicit memory-review verdict gate, and requests a targeted post-accept feature recheck that preserves unrelated features and conflict flags.
diff --git a/memory/index.md b/memory/index.md
index 16d67f2..33780d0 100644
--- a/memory/index.md
+++ b/memory/index.md
@@ -10,6 +10,7 @@ last_memorized_commit: c5b75e3e4b034bc80a1600e57249e162639d32a8
 * [Design](design.md) - A compact dark developer control plane with clear workflow state.
 * [Backlog](backlog/index.md) - Saved ideas and bugs outside active plan features.
 * [Settings](settings.md) - Project settings (auto mode, models, git flow).
+* [Features](features/) - One document per implementation feature.
 * [Architecture](/architecture/) - How the system is structured.
 * [Decisions](/decisions/) - Durable product and implementation choices agents should preserve.
 * [Patterns & Conventions](/patterns/) - How code and workflows are written in this repo.
diff --git a/memory/log.md b/memory/log.md
index 3b27155..de930f1 100644
--- a/memory/log.md
+++ b/memory/log.md
@@ -1,5 +1,96 @@
 # iterator update log
 
+## 2026-07-20
+* **Backlog**: select remove-text-erferences-to-claude (selected).
+* **Backlog**: select red-mode-tests-should-show-the-real-test-code-for-verification-2 (selected).
+* **Backlog**: select red-mode-tests-should-show-the-real-test-code-for-verification (selected).
+* **Backlog**: create remove-text-erferences-to-claude.
+* **Retirement**: Plan "Streamline backlog planning and knowledge actions" condensed into [Streamline backlog planning and knowledge actions](/decisions/streamline-backlog-planning-and-knowledge-actions.md).
+* **Plan review**: Whole-plan review recorded (agent).
+* **Review**: Reviewed [Work conflict memory resolution](/features/work-conflict-memory-resolution.md); approved (agent).
+* **Review**: Accepted [Work conflict memory resolution](/features/work-conflict-memory-resolution.md) (committed as feature(work-conflict-memory-resolution)).
+* **Implementation**: Committed feature(work-conflict-memory-resolution) on branch iterator/safe-role-model-handoff; awaiting review.
+* **Review**: Reviewed [Focused memory review changes](/features/memory-review-change-focus.md); approved (agent).
+* **Review**: Accepted [Focused memory review changes](/features/memory-review-change-focus.md) (committed as feature(memory-review-change-focus)).
+* **Implementation**: Committed feature(memory-review-change-focus) on branch iterator/safe-role-model-handoff; awaiting review.
+* **Review**: Reviewed [Focused memory review changes](/features/memory-review-change-focus.md); changes (agent).
+* **Implementation**: Committed feature(memory-review-change-focus) on branch iterator/safe-role-model-handoff; awaiting review.
+* **Review**: Reviewed [Full-plan fast track](/features/full-plan-fast-track.md); approved (agent).
+* **Review**: Accepted [Full-plan fast track](/features/full-plan-fast-track.md) (committed as feature(full-plan-fast-track)).
+* **Implementation**: Committed feature(full-plan-fast-track) on branch iterator/safe-role-model-handoff; awaiting review.
+* **Review**: Reviewed [Backlog filtering and bulk selection](/features/backlog-filter-and-bulk-select.md); approved (agent).
+* **Review**: Accepted [Backlog filtering and bulk selection](/features/backlog-filter-and-bulk-select.md) (committed as feature(backlog-filter-and-bulk-select)).
+* **Implementation**: Committed feature(backlog-filter-and-bulk-select) on branch iterator/safe-role-model-handoff; awaiting review.
+* **Review**: Reviewed [Backlog filtering and bulk selection](/features/backlog-filter-and-bulk-select.md); changes (agent).
+* **Implementation**: Committed feature(backlog-filter-and-bulk-select) on branch iterator/safe-role-model-handoff; awaiting review.
+* **Review**: Reviewed [Reliable backlog saves](/features/backlog-save-during-work.md); approved (agent).
+* **Review**: Accepted [Reliable backlog saves](/features/backlog-save-during-work.md) (committed as feature(backlog-save-during-work)).
+* **Implementation**: Committed feature(backlog-save-during-work) on branch iterator/safe-role-model-handoff; awaiting review.
+* **Review**: Reviewed [Backlog file mentions](/features/backlog-file-mentions.md); approved (agent).
+* **Review**: Accepted [Backlog file mentions](/features/backlog-file-mentions.md) (committed as feature(backlog-file-mentions)).
+* **Implementation**: Committed feature(backlog-file-mentions) on branch iterator/safe-role-model-handoff; awaiting review.
+* **Update**: Applied 6 feature adjustment(s).
+* **Creation**: 6 feature(s) written.
+* **Creation**: Plan "Streamline backlog planning and knowledge actions" approved on branch iterator/safe-role-model-handoff. Consumed 6 selected backlog candidate(s).
+* **Backlog**: select when-adding-a-plan-just-create-the-features (selected).
+* **Backlog**: select like-in-plan-allow-in-the-idea-baclkog-inputs-the-same-file-seleciton-logic-with (selected).
+* **Backlog**: select memroy-update-needs-foxus (selected).
+* **Backlog**: select save-candidate-should-alsways-work-as-its-a-save-to-file (selected).
+* **Backlog**: select in-work-ifeature-warning-that-it-conflics-with-a-memory-add-a-accept-button (selected).
+* **Backlog**: select idea-backlog-needs-filter-by-feature-and-bug-and-selecte-deselect-all-buttons (selected).
+* **Backlog**: create idea-backlog-needs-filter-by-feature-and-bug-and-selecte-deselect-all-buttons.
+* **Backlog**: select in-work-ifeature-warning-that-it-conflics-with-a-memory-add-a-accept-button (deselected).
+* **Backlog**: select in-work-ifeature-warning-that-it-conflics-with-a-memory-add-a-accept-button (selected).
+* **Retirement**: Plan "Focus feature execution and dashboard ownership" condensed into [Focus feature execution and dashboard ownership](/decisions/focus-feature-execution-and-dashboard-ownership.md).
+* **Plan review**: Whole-plan review recorded (agent).
+* **Review**: Reviewed [Visible red-test handoff](/features/visible-red-test-handoff.md); approved.
+* **Review**: Accepted [Visible red-test handoff](/features/visible-red-test-handoff.md) (committed as feature(visible-red-test-handoff)).
+* **Implementation**: Committed feature(visible-red-test-handoff) on branch iterator/safe-role-model-handoff; awaiting review.
+* **Backlog**: create in-work-ifeature-warning-that-it-conflics-with-a-memory-add-a-accept-button.
+* **Review**: Reviewed [Settings dashboard modal](/features/settings-dashboard-modal.md); approved.
+* **Review**: Accepted [Settings dashboard modal](/features/settings-dashboard-modal.md) (committed as feature(settings-dashboard-modal)).
+* **Update**: Memorized [Settings modal preserves originating dashboard context](/decisions/settings-close-returns-to-work.md).
+* **Implementation**: Committed feature(settings-dashboard-modal) on branch iterator/safe-role-model-handoff; awaiting review.
+* **Review**: Reviewed [Active plan workspace](/features/active-plan-workspace.md); approved (agent).
+* **Review**: Accepted [Active plan workspace](/features/active-plan-workspace.md) (committed as feature(active-plan-workspace)).
+* **Implementation**: Committed feature(active-plan-workspace) on branch iterator/safe-role-model-handoff; awaiting review.
+* **Review**: Reviewed [Active plan workspace](/features/active-plan-workspace.md); changes (agent).
+* **Update**: Memorized [Work owns active plan context and lifecycle](/decisions/review-navigation-and-work-context.md).
+* **Implementation**: Committed feature(active-plan-workspace) on branch iterator/safe-role-model-handoff; awaiting review.
+* **Review**: Reviewed [Fresh implementation session](/features/fresh-implementation-session.md); approved (agent).
+* **Review**: Accepted [Fresh implementation session](/features/fresh-implementation-session.md) (committed as feature(fresh-implementation-session)).
+* **Implementation**: Committed feature(fresh-implementation-session) on branch iterator/safe-role-model-handoff; awaiting review.
+* **Review**: Reviewed [Fresh implementation session](/features/fresh-implementation-session.md); changes (agent).
+* **Implementation**: Committed feature(fresh-implementation-session) on branch iterator/safe-role-model-handoff; awaiting review.
+* **Update**: Applied 4 feature adjustment(s).
+* **Creation**: 4 feature(s) written.
+* **Creation**: Plan "Focus feature execution and dashboard ownership" approved on branch iterator/safe-role-model-handoff. Consumed 5 selected backlog candidate(s).
+* **Backlog**: select openeing-setting-should-be-an-overaly (selected).
+* **Backlog**: select it-looks-as-if-the-agent-isnt-using-new-before-a-new-feature-is-implemented-in-either-auto-or-manual-mode (selected).
+* **Backlog**: select move-all-active-plan-related-infos-to-work (selected).
+* **Backlog**: select after-feature-show-work-tab (selected).
+* **Backlog**: select after-adding-tests-its-not-clear-that-they-are-there (selected).
+* **Retirement**: Plan "Fix post-test role model credential corruption" condensed into [Safely restore configured Iterator role models](/decisions/safe-role-model-restoration.md).
+* **Backlog**: create save-candidate-should-alsways-work-as-its-a-save-to-file.
+* **Backlog**: create memroy-update-needs-foxus.
+* **Update**: Memorized [Apply role models to manual turns and reset stale runtime state](/decisions/manual-role-models-and-runtime-reset.md).
+* **Review**: Accepted [Safe role-model handoff](/features/safe-role-model-handoff.md) (committed as feature(safe-role-model-handoff)).
+* **Implementation**: Committed feature(safe-role-model-handoff) on branch iterator/safe-role-model-handoff; awaiting review.
+* **Backlog**: create red-mode-tests-should-show-the-real-test-code-for-verification-2.
+* **Backlog**: create red-mode-tests-should-show-the-real-test-code-for-verification.
+* **Update**: Applied 1 feature adjustment(s).
+* **Creation**: 1 feature(s) written.
+* **Creation**: Plan "Fix post-test role model credential corruption" approved on branch main. Consumed 1 selected backlog candidate(s).
+* **Backlog**: select after-red-mode-test-creation-iterator-doesnt-work (selected).
+* **Backlog**: create after-red-mode-test-creation-iterator-doesnt-work.
+* **Backlog**: create after-adding-tests-its-not-clear-that-they-are-there.
+* **Backlog**: create openeing-setting-should-be-an-overaly.
+* **Backlog**: create move-all-active-plan-related-infos-to-work.
+* **Backlog**: create after-feature-show-work-tab.
+* **Backlog**: create like-in-plan-allow-in-the-idea-baclkog-inputs-the-same-file-seleciton-logic-with.
+* **Backlog**: create when-adding-a-plan-just-create-the-features.
+* **Backlog**: create it-looks-as-if-the-agent-isnt-using-new-before-a-new-feature-is-implemented-in-either-auto-or-manual-mode.
+
 ## 2026-07-19
 * **Retirement**: Plan "Improve memory relevance, usage costs, and dashboard recovery" condensed into [Memory relevance, usage costs, and dashboard recovery](/decisions/memory-relevance-usage-and-dashboard-recovery.md).
 
diff --git a/memory/state.md b/memory/state.md
index d61f6f9..88106d4 100644
--- a/memory/state.md
+++ b/memory/state.md
@@ -3,12 +3,12 @@ type: State
 title: Runtime state
 description: Machine-owned iterator flow state — never hand-edited.
 mode: manual
-paused: false
+paused: true
 phase: done
 active_feature: null
-strikes: "{\"onboard-planless-projects\":1,\"reliable-work-blocker\":1}"
+strikes: "{\"backlog-filter-and-bulk-select\":1,\"memory-review-change-focus\":1}"
 escalation: null
-timestamp: 2026-07-18T07:58:05.493Z
+timestamp: 2026-07-20T16:02:54.672Z
 ---
 
 Runtime flow state; read via gather, written only by the state op.
diff --git a/memory/usage.md b/memory/usage.md
index 0145f41..21c0088 100644
--- a/memory/usage.md
+++ b/memory/usage.md
@@ -1,23 +1,30 @@
 ---
 type: Usage
 title: Token usage
-description: Per-step model/token ledger for the active plan — written only by the usage op.
-totals: "{\"steps\":{\"hub\":{\"openai-codex/gpt-5.6-terra\":{\"input\":261034,\"output\":1116,\"cacheRead\":1795072,\"cacheWrite\":0,\"turns\":8}}},\"features\":{\"retire-plan\":{\"input\":261034,\"output\":1116,\"cacheRead\":1795072,\"cacheWrite\":0,\"turns\":8}}}"
-timestamp: 2026-07-19T07:33:07.539Z
+description: Per-step model/token ledger and optional project-owned pricing for the active plan — written only by the usage op.
+totals: "{\"steps\":{\"hub\":{\"openai-codex/gpt-5.6-terra\":{\"input\":38403,\"output\":1157,\"cacheRead\":217088,\"cacheWrite\":0,\"turns\":8}},\"plan\":{\"openai-codex/gpt-5.6-sol\":{\"input\":36212,\"output\":221,\"cacheRead\":0,\"cacheWrite\":0,\"turns\":2}}},\"features\":{\"retire-plan\":{\"input\":38403,\"output\":1157,\"cacheRead\":217088,\"cacheWrite\":0,\"turns\":8}},\"featureModels\":{\"retire-plan\":{\"openai-codex/gpt-5.6-terra\":{\"input\":38403,\"output\":1157,\"cacheRead\":217088,\"cacheWrite\":0,\"turns\":8}}}}"
+prices: "{}"
+timestamp: 2026-07-20T16:04:49.832Z
 ---
 
 # Usage
 
 ## hub
 
-| model | input | output | cache read | cache write | turns |
-| --- | ---: | ---: | ---: | ---: | ---: |
-| openai-codex/gpt-5.6-terra | 261034 | 1116 | 1795072 | 0 | 8 |
+| model | input | output | cache read | cache write | turns | cost |
+| --- | ---: | ---: | ---: | ---: | ---: | ---: |
+| openai-codex/gpt-5.6-terra | 38403 | 1157 | 217088 | 0 | 8 | — |
+
+## plan
+
+| model | input | output | cache read | cache write | turns | cost |
+| --- | ---: | ---: | ---: | ---: | ---: | ---: |
+| openai-codex/gpt-5.6-sol | 36212 | 221 | 0 | 0 | 2 | — |
 
 ## Per feature
 
-| feature | input | output | cache read | cache write | turns |
-| --- | ---: | ---: | ---: | ---: | ---: |
-| retire-plan | 261034 | 1116 | 1795072 | 0 | 8 |
+| feature | input | output | cache read | cache write | turns | cost |
+| --- | ---: | ---: | ---: | ---: | ---: | ---: |
+| retire-plan | 38403 | 1157 | 217088 | 0 | 8 | — |
 
-Total: 261034 in / 1116 out / 1795072 cache-read / 0 cache-write over 8 turns.
+Total: 74615 in / 1378 out / 217088 cache-read / 0 cache-write over 10 turns. Cost unavailable: add every used model rate.
diff --git a/skills/iterator-implement/SKILL.md b/skills/iterator-implement/SKILL.md
index 91b8648..cdd177c 100644
--- a/skills/iterator-implement/SKILL.md
+++ b/skills/iterator-implement/SKILL.md
@@ -14,7 +14,13 @@ commits its own files, unrelated working-tree churn never pollutes a review,
 and several independent features can be implemented back to back and reviewed
 afterwards.
 
-**pi mode:** see `<skill-dir>/../iterator/PI.md`.
+**pi mode:** see `<skill-dir>/../iterator/PI.md`. The friendly
+`/iterator-implement` and `/iterator-next` commands create a replacement Pi
+session before running the feature skill. That session receives only the named
+implementation command and rebuilds its context from the deterministic
+implement gather contract. Dashboard and auto-mode implementation dispatches
+use the same command. A direct `/skill:iterator-implement` invocation remains
+available for harnesses that already started the dedicated session.
 
 **Claude Code mode:** Use gather/write directly instead of the Pi dashboard.
 Implement only `next`, commit it through `commit-feature`, and report it as
diff --git a/skills/iterator/lib/gather.mjs b/skills/iterator/lib/gather.mjs
index 4394ada..f13f78e 100644
--- a/skills/iterator/lib/gather.mjs
+++ b/skills/iterator/lib/gather.mjs
@@ -395,6 +395,7 @@ export function gather(startDir) {
 	];
 
 	const trailerMap = featureCommitMap(b.root);
+	const conceptsById = new Map(loadConcepts(b.memDir).map((c) => [c.id, c]));
 	// Readiness is computed here, once, server-side — views only render it.
 	const ready = readiness(b.features, b.settings);
 	const features = b.features.map((c) => {
@@ -403,6 +404,17 @@ export function gather(startDir) {
 		const hasDiff = diffFiles.some((f) => globs.some((re) => re.test(f)));
 		const recorded = Array.isArray(c.fm.commits) && c.fm.commits.length > 0;
 		const hasCommits = recorded || trailerMap.has(c.slug);
+		const tests = listy(c.fm.tests);
+		const conflicts = featureConflicts(c.fm).map((conflict) => {
+			const decision = conceptsById.get(conflict.decision);
+			return {
+				...conflict,
+				title: decision?.title || conflict.decision,
+				description: decision?.description || "",
+				files: decision?.files || [],
+				decisionExists: Boolean(decision),
+			};
+		});
 		return {
 			name: c.slug,
 			title: c.fm.title || c.slug,
@@ -410,11 +422,13 @@ export function gather(startDir) {
 			status: c.fm.status || "pending",
 			size: c.fm.size || "small",
 			testsStatus: c.fm.tests_status || "none",
+			tests,
+			testCount: tests.length,
 			dependsOn: listy(c.fm.depends_on),
 			...ready.get(c.slug),
 			hasDiff,
 			hasCommits,
-			conflicts: featureConflicts(c.fm).length,
+			conflicts,
 		};
 	});
 
diff --git a/skills/iterator/lib/ui.mjs b/skills/iterator/lib/ui.mjs
index c4eac2a..0cc346b 100644
--- a/skills/iterator/lib/ui.mjs
+++ b/skills/iterator/lib/ui.mjs
@@ -217,6 +217,10 @@ async function post(payload, okMsg, options){
 function mdToHtml(src){
   if(!src || !src.trim()) return '<span class="empty">Empty.</span>';
   const escc = s => s.replace(/&/g,'&amp;').replace(/</g,'&lt;').replace(/>/g,'&gt;');
+  // inline() already escaped &, <, and > before link matching; only quotes
+  // remain unsafe in the captured href. Escaping the whole URL again would
+  // turn &amp; into &amp;amp; and corrupt query parameters.
+  const attr = s => s.replace(/"/g,'&quot;').replace(/'/g,'&#39;');
   const inline = s => escc(s)
     .replace(/\`([^\`]+)\`/g, (m,c)=>'<code>'+c+'</code>')
     .replace(/\\*\\*([^*]+)\\*\\*/g, '<strong>$1</strong>')
@@ -224,7 +228,7 @@ function mdToHtml(src){
     .replace(/\\[([^\\]]+)\\]\\(([^)]+)\\)/g, function(m,t,u){
       // Only linkify http(s)/mailto/relative targets — never javascript: etc.
       return /^(https?:|mailto:|[\\/.#])/i.test(u)
-        ? '<a href="'+u+'" target="_blank" rel="noopener">'+t+'</a>' : m;
+        ? '<a href="'+attr(u)+'" target="_blank" rel="noopener">'+t+'</a>' : m;
     });
   const lines = src.split('\\n');
   let html='', i=0, inList=false, listTag='';
@@ -255,6 +259,50 @@ function mdToHtml(src){
   closeList();
   return html;
 }
+
+// Split markdown into heading-owned blocks, preserving fenced code verbatim.
+// Used by memory review to focus changed sections without trusting raw HTML.
+function markdownBlocks(src){
+  var lines = String(src == null ? '' : src).replace(/\\r/g, '').split('\\n');
+  var blocks = [], current = [], key = '$intro', labels = {}, fenced = false;
+  function push(){
+    var text = current.join('\\n').trim();
+    if(text) blocks.push({ key:key, text:text });
+    current = [];
+  }
+  lines.forEach(function(line){
+    var trim = line.trim();
+    if(trim.startsWith('\`\`\`') || trim.startsWith('~~~')) fenced = !fenced;
+    var match = fenced ? null : line.match(/^(#{1,6})\\s+(.+)$/);
+    if(match){
+      push();
+      var base = match[2].trim().toLowerCase();
+      labels[base] = (labels[base] || 0) + 1;
+      key = base + '#' + labels[base];
+    }
+    current.push(line);
+  });
+  push();
+  return blocks;
+}
+function markdownChangeHtml(next, previous, isCreate){
+  var before = markdownBlocks(previous);
+  var after = markdownBlocks(next);
+  var old = new Map(before.map((block)=> [block.key, block]));
+  var seen = new Set();
+  var html = after.map((block)=> {
+    var prior = old.get(block.key);
+    seen.add(block.key);
+    var kind = isCreate || !prior ? 'added' : (prior.text === block.text ? 'unchanged' : 'modified');
+    var label = kind === 'added' ? 'New section' : (kind === 'modified' ? 'Changed section' : 'Unchanged');
+    return '<section class="change-block '+kind+'"><div class="change-label">'+label+'</div>'+mdToHtml(block.text)+'</section>';
+  }).join('');
+  before.forEach((block)=> {
+    if(seen.has(block.key)) return;
+    html += '<section class="change-block removed"><div class="change-label">Removed section</div>'+mdToHtml(block.text)+'</section>';
+  });
+  return html || mdToHtml(next);
+}
 `;
 
 /* ------------------------------------------------------------------ *
diff --git a/skills/iterator/lib/views/hub.mjs b/skills/iterator/lib/views/hub.mjs
index 90dad21..7d45781 100644
--- a/skills/iterator/lib/views/hub.mjs
+++ b/skills/iterator/lib/views/hub.mjs
@@ -2,9 +2,10 @@
  * iterator: Work dashboard UI on the shared shell (../ui.mjs,
  * ../server.mjs). The execution surface of the flow: the active plan's
  * progress, dependency graph, and feature cards with their Test / Implement /
- * Review actions. Planning owns backlog and plan-lifecycle controls; active
- * plan context and feature management live here. Both surfaces render from
- * the same gather payload so they can never disagree about state.
+ * Review actions. Work owns active-plan progress, lifecycle controls, and
+ * feature management; Planning owns backlog, future planning, and archives.
+ * Both surfaces render from the same gather payload so they cannot disagree
+ * about state.
  *
  *   input:  { step:"hub", branch,
  *             plan: { title, status } | null,      // null = no bundle yet
@@ -14,13 +15,15 @@
  *             dirty: { count, files },             // working-tree files outside the bundle
  *             features: [ { name, title, description, status, size,
  *                         testsStatus,                  // none | red | green
+ *                         tests, testCount,              // recorded red/green test targets
  *                         dependsOn, ready, waitingOn,  // server-computed readiness
  *                         hasDiff, hasCommits,
- *                         conflicts } ] }               // # of flagged decision conflicts
+ *                         conflicts } ] }               // flagged decisions + notes/anchors
  *             retired: [ { name, title, created } ]   // archived plans, newest first
  *   output: one JSON line to stdout —
- *     { type:"action", action:"planning"|"test"|"implement"|"review"|"review-all"|"implement-wave"|"auto-implement"
- *       |"cancel-feature"|"escalation-restart"|"escalation-guide",
+ *     { type:"action", action:"planning"|"plan"|"feature"|"test"|"implement"|"review"|"review-all"
+ *       |"review-plan"|"retire"|"implement-wave"|"auto-implement"|"cancel-plan"|"cancel-feature"
+ * *       |"escalation-restart"|"escalation-guide"|"resolve-memory-conflict",
  *       feature:"<slug>"|null,
  *       prompt:"<guidance>"|null }                 // escalation-guide only
  *     plus the shared { type:"cancel" } / { type:"timeout" }.
@@ -35,6 +38,8 @@ const CSS = `
 .escalation .er{font-size:var(--fs-sm);white-space:pre-wrap}
 .escalation .em{font-size:var(--fs-xs);color:var(--text-muted)}
 .escalation textarea{width:100%;min-height:56px;resize:vertical;background:var(--bg);border:1px solid var(--border);border-radius:var(--radius-sm);padding:8px 10px;color:var(--text);font:inherit;font-size:var(--fs-sm)}
+.conflict-list{display:grid;gap:var(--sp-2);margin:8px 0 10px}.conflict-item{display:flex;align-items:flex-start;justify-content:space-between;gap:var(--sp-3);padding:8px 10px;background:var(--bg-red);border-left:3px solid var(--dot-red);border-radius:0 var(--radius-sm) var(--radius-sm) 0}.conflict-copy{font-size:var(--fs-xs);line-height:1.45}.conflict-copy strong{display:block;color:var(--del-fg)}.conflict-copy code{font-size:11px}.conflict-action{flex-shrink:0}
+@media(max-width:640px){.conflict-item{flex-direction:column}.conflict-action{width:100%}}
 `;
 
 const BODY = `
@@ -99,8 +104,8 @@ function render(){
     w.appendChild(box);
   }
 
-  // Plan bar: progress + the execution controls. Lifecycle buttons (revise,
-  // feature, review-plan, retire, cancel) live on the Planning surface.
+  // Work keeps active-plan progress, lifecycle, and execution controls in one
+  // plan bar; Planning remains a staging and archive surface.
   const done = (D.progress&&D.progress.done)||0, total = (D.progress&&D.progress.total)||CH.length;
   const bar = document.createElement('div');
   bar.className = 'planbar';
@@ -110,6 +115,52 @@ function render(){
       ? '<span class="pcount">'+done+' / '+total+' features done</span>'+
         '<div class="pbar"><div style="width:'+Math.round(done/total*100)+'%"></div></div>'
       : '<span class="pcount">not broken into features yet</span>');
+  // Active-plan lifecycle belongs with its progress and feature work. Every
+  // condition is the server-derived stage; this view never infers it.
+  const revise = document.createElement('button');
+  revise.className = 'act';
+  revise.textContent = 'Revise plan';
+  revise.addEventListener('click', () => action('plan', null, 'Starting /iterator-plan'));
+  bar.appendChild(revise);
+  const refeature = document.createElement('button');
+  refeature.className = D.stage==='needs-features' ? 'act primary-act' : 'act';
+  refeature.textContent = D.stage==='needs-features' ? 'Feature the plan' : 'Re-feature';
+  refeature.title = D.stage==='needs-features'
+    ? 'Next step: break the approved plan into small, dependency-ordered features (/iterator-feature)'
+    : 'Redraw the feature set from the plan (/iterator-feature)';
+  refeature.addEventListener('click', () => action('feature', null, 'Starting /iterator-feature'));
+  bar.appendChild(refeature);
+  if(D.stage==='awaiting-plan-review' || D.stage==='retirable'){
+    const reviewPlan = document.createElement('button');
+    reviewPlan.className = 'act primary-act';
+    reviewPlan.textContent = D.plan.planReviewed ? 'Re-review plan' : 'Review plan';
+    reviewPlan.title = D.plan.planReviewed
+      ? 'Plan reviewed '+D.plan.planReviewed+' \\u2014 run the whole-plan review again'
+      : 'Review all changes and commits against the plan\\u2019s goals and decisions';
+    reviewPlan.addEventListener('click', () => action('review-plan', null, 'Starting /iterator-review-plan'));
+    bar.appendChild(reviewPlan);
+  }
+  if(D.stage==='retirable'){
+    const retire = document.createElement('button');
+    retire.className = 'act primary-act';
+    retire.textContent = 'Retire plan';
+    retire.title = 'Condense the finished plan into a decisions/ memory and archive its features';
+    confirmButton(retire, 'Retires the plan \\u2014 click again', () =>
+      action('retire', null, 'Starting plan retirement'));
+    bar.appendChild(retire);
+  }
+  const cancelPlan = document.createElement('button');
+  cancelPlan.className = 'act danger';
+  cancelPlan.textContent = 'Cancel plan';
+  const dirtyWarn = (D.dirty && D.dirty.count)
+    ? D.dirty.count + ' uncommitted file' + (D.dirty.count!==1?'s':'') + ' + '
+    : '';
+  cancelPlan.title = 'Abandon this plan: archive it and DELETE its branch/worktree'
+    + (dirtyWarn ? ' \\u2014 \\u26a0 ' + dirtyWarn + 'unmerged commits will be lost' : '');
+  confirmButton(cancelPlan, '\\u26a0 Deletes ' + dirtyWarn + 'branch \\u2014 click again', () =>
+    action('cancel-plan', null, 'Cancelling plan'));
+  bar.appendChild(cancelPlan);
+
   // Auto mode: once the feature set is approved (pending features exist), the
   // whole test → implement → review loop can run agent-driven.
   const readyWave = Array.isArray(D.readyWave) ? D.readyWave : [];
@@ -176,7 +227,13 @@ function makeCard(c){
   const ready = c.ready !== false;
   const draft = c.status==='draft';
   const implemented = c.status==='implemented';
+  const conflictDetails = Array.isArray(c.conflicts) ? c.conflicts : [];
+  const conflictCount = conflictDetails.length || Number(c.conflicts) || 0;
   const icon = '<i class="st '+(c.status==='done'?'done':draft?'draft':implemented?'implemented':'pending')+'"></i>';
+  const redTarget = c.testsStatus==='red'
+    ? '<div class="deps"><strong>Committed red tests — implementation target:</strong> '+
+      ((c.tests&&c.tests.length) ? c.tests.map(t=>'<code>'+esc(t)+'</code>').join(' ') : esc(String(c.testCount||0))+' test'+(c.testCount===1?'':'s'))+'</div>'
+    : '';
   card.innerHTML =
     '<div class="ch"><span class="cn">'+icon+esc(c.title||c.name)+'</span>'+
       '<span class="chip cmut">'+esc(c.name)+'</span>'+
@@ -184,16 +241,38 @@ function makeCard(c){
       (implemented?'<span class="chip cy">implemented \\u2014 awaiting review</span>':'')+
       (implemented&&!c.hasDiff&&!c.hasCommits?'<span class="chip cr" title="Marked implemented, but review would find nothing: no working-tree diff and no recorded commits. Usual causes: the work was committed outside the accept flow (no Feature: trailer) or landed in a different checkout (plan worktree vs main). Check git log for its files, or Restart the feature.">\\u26a0 no recorded changes</span>':'')+
       sizeChip(c)+testBadge(c)+
-      (c.conflicts?'<span class="chip cr" title="This feature contradicts a project decision — check its Decision conflicts section">\\u26a0 '+c.conflicts+' decision conflict'+(c.conflicts!==1?'s':'')+'</span>':'')+'</div>'+
-    '<div class="cdesc">'+esc(c.description||'')+'</div>'+
+      (conflictCount?'<span class="chip cr" title="This feature contradicts a project decision — review it before implementation">\\u26a0 '+conflictCount+' decision conflict'+(conflictCount!==1?'s':'')+'</span>':'')+'</div>'+
+    '<div class="cdesc">'+esc(c.description||'')+'</div>'+redTarget+
     ((c.dependsOn&&c.dependsOn.length)?'<div class="deps">depends on '+c.dependsOn.map(d=>'<code>'+esc(d)+'</code>').join(' ')+'</div>':'');
 
+  if(conflictDetails.length){
+    const list = document.createElement('div'); list.className = 'conflict-list';
+    conflictDetails.forEach(conflict => {
+      const row = document.createElement('div'); row.className = 'conflict-item';
+      const copy = document.createElement('div'); copy.className = 'conflict-copy';
+      const title = document.createElement('strong'); title.textContent = conflict.title || conflict.decision;
+      const note = document.createElement('div'); note.textContent = conflict.note || conflict.description || 'Recorded decision conflict';
+      const id = document.createElement('code'); id.textContent = conflict.decision;
+      copy.append(title, note, id);
+      const resolve = document.createElement('button'); resolve.className = 'act conflict-action'; resolve.type = 'button'; resolve.textContent = 'Review decision';
+      resolve.title = 'Draft a decision update, require memory review, then re-check only this feature conflict';
+      resolve.addEventListener('click', () => post({
+        type:'action', action:'resolve-memory-conflict', feature:c.name,
+        target:conflict.decision, prompt:conflict.note || null,
+        anchors:Array.isArray(conflict.files) ? conflict.files : []
+      }, 'Starting reviewed decision update'));
+      row.append(copy, resolve); list.appendChild(row);
+    });
+    card.appendChild(list);
+  }
+
   const btns = document.createElement('div');
   btns.className = 'btns';
 
   const impl = document.createElement('button');
   impl.className = 'act primary-act';
-  impl.textContent = 'Implement';
+  impl.textContent = c.testsStatus==='red' ? 'Drive tests green' : 'Implement';
+  if(c.testsStatus==='red') impl.title = 'Run the committed, intentionally failing tests and make them pass';
   if(c.status==='done'){ impl.disabled = true; impl.title = 'Already done'; }
   else if(implemented){ impl.disabled = true; impl.title = 'Implemented — review it'; }
   else if(draft){ impl.disabled = true; impl.title = 'Draft — accept the feature set first (Re-feature on the Planning tab)'; }
@@ -205,11 +284,14 @@ function makeCard(c){
 
   const test = document.createElement('button');
   test.className = 'act';
-  test.textContent = c.status==='pending' ? 'Test (red)' : 'Test';
-  test.title = c.status==='pending'
-    ? 'Write failing tests from the feature contract before implementing'
-    : 'Write passing tests against the implemented feature';
-  if(draft){ test.disabled = true; test.title = 'Draft — accept the feature set first'; }
+  test.textContent = c.testsStatus==='red' ? 'Tests committed (red)' : (c.status==='pending' ? 'Test (red)' : 'Test');
+  test.title = c.testsStatus==='red'
+    ? 'Committed red tests are the implementation target — drive them green'
+    : c.status==='pending'
+      ? 'Write failing tests from the feature contract before implementing'
+      : 'Write passing tests against the implemented feature';
+  if(c.testsStatus==='red'){ test.disabled = true; }
+  else if(draft){ test.disabled = true; test.title = 'Draft — accept the feature set first'; }
   test.addEventListener('click', () => action('test', c.name, 'Starting /iterator-test'));
 
   const rev = document.createElement('button');
diff --git a/skills/iterator/lib/views/memory-review.mjs b/skills/iterator/lib/views/memory-review.mjs
index 8ca0d29..8a1642f 100644
--- a/skills/iterator/lib/views/memory-review.mjs
+++ b/skills/iterator/lib/views/memory-review.mjs
@@ -22,74 +22,98 @@
  * change nothing) | keep (leave the existing file untouched) | delete (remove
  * the existing concept file).
  */
-import { renderPage, escHtml } from '../ui.mjs';
+import { renderPage, escHtml } from "../ui.mjs";
 
 const ACTION_BADGE = {
-  create: ['new', 'NEW'],
-  update: ['upd', 'UPDATE'],
-  delete: ['del', 'DELETE'],
-  keep: ['keep', 'KEEP'],
+	create: ["new", "NEW"],
+	update: ["upd", "UPDATE"],
+	delete: ["del", "DELETE"],
+	keep: ["keep", "KEEP"],
 };
 
 function banner(data, memories) {
-  const mode = data.mode;
-  const staleCount = memories.filter(m => m.stale).length;
-  const conflictCount = memories.filter(m => m.conflict).length;
-  const draftCount = memories.filter(m => m.action !== 'keep').length;
-  if (mode === 'init') {
-    return `Drafted <strong>${memories.length}</strong> memories from codebase analysis — review, comment, accept.`;
-  }
-  if (mode === 'consolidate') {
-    return `Reviewing <strong>${memories.length}</strong> existing memories`
-      + (staleCount ? ` — <span class="amber"><strong>${staleCount}</strong> stale</span>` : ' — none stale')
-      + '.';
-  }
-  const commitCount = Number(data.commitCount) || 0;
-  const draftLabel = draftCount === 1 ? 'draft' : 'drafts';
-  const commitLabel = commitCount === 1 ? 'commit' : 'commits';
-  const conflictLabel = conflictCount === 1 ? 'conflict' : 'conflicts';
-  const conflictHtml = conflictCount
-    ? ` — <span class="red"><strong>${conflictCount}</strong> ${conflictLabel}</span>`
-    : ' — no conflicts';
-  return `<strong>${draftCount}</strong> ${draftLabel} from `
-    + `<strong>${commitCount || '?'}</strong> ${commitLabel}` + conflictHtml + '.';
+	const mode = data.mode;
+	const staleCount = memories.filter((m) => m.stale).length;
+	const conflictCount = memories.filter((m) => m.conflict).length;
+	const draftCount = memories.filter((m) => m.action !== "keep").length;
+	if (mode === "init") {
+		return `Drafted <strong>${memories.length}</strong> memories from codebase analysis — review, comment, accept.`;
+	}
+	if (mode === "consolidate") {
+		return (
+			`Reviewing <strong>${memories.length}</strong> existing memories` +
+			(staleCount
+				? ` — <span class="amber"><strong>${staleCount}</strong> stale</span>`
+				: " — none stale") +
+			"."
+		);
+	}
+	const commitCount = Number(data.commitCount) || 0;
+	const draftLabel = draftCount === 1 ? "draft" : "drafts";
+	const commitLabel = commitCount === 1 ? "commit" : "commits";
+	const conflictLabel = conflictCount === 1 ? "conflict" : "conflicts";
+	const conflictHtml = conflictCount
+		? ` — <span class="red"><strong>${conflictCount}</strong> ${conflictLabel}</span>`
+		: " — no conflicts";
+	return (
+		`<strong>${draftCount}</strong> ${draftLabel} from ` +
+		`<strong>${commitCount || "?"}</strong> ${commitLabel}` +
+		conflictHtml +
+		"."
+	);
 }
 
 function verdictButtons(m) {
-  const btn = (verdict, label) =>
-    `<button type="button" data-verdict="${verdict}">${label}</button>`;
-  if (m.action === 'keep') return btn('keep', 'Keep') + btn('delete', 'Delete');
-  if (m.action === 'delete') return btn('delete', 'Delete') + btn('keep', 'Keep');
-  return btn('accept', 'Accept') + btn('reject', 'Reject');
+	const btn = (verdict, label) =>
+		`<button type="button" data-verdict="${verdict}">${label}</button>`;
+	if (m.action === "keep") return btn("keep", "Keep") + btn("delete", "Delete");
+	if (m.action === "delete")
+		return btn("delete", "Delete") + btn("keep", "Keep");
+	return btn("accept", "Accept") + btn("reject", "Reject");
 }
 
 function cardHtml(m) {
-  const [cls, label] = ACTION_BADGE[m.action] || ACTION_BADGE.keep;
-  const tags = Array.isArray(m.tags) && m.tags.length
-    ? ' · ' + m.tags.map(t => `<span class="tag">${escHtml(t)}</span>`).join(' ')
-    : '';
-  const commits = Array.isArray(m.sourceCommits) && m.sourceCommits.length
-    ? ` · <span class="commits">${escHtml(m.sourceCommits.map(c => String(c).slice(0, 7)).join(', '))}</span>`
-    : '';
-  const showExisting = m.existingBody !== null && m.existingBody !== undefined
-    && ['update', 'delete', 'keep'].includes(m.action);
-  return `<article class="card" data-id="${escHtml(m.id)}">
+	const [cls, label] = ACTION_BADGE[m.action] || ACTION_BADGE.keep;
+	const tags =
+		Array.isArray(m.tags) && m.tags.length
+			? " · " +
+				m.tags.map((t) => `<span class="tag">${escHtml(t)}</span>`).join(" ")
+			: "";
+	const commits =
+		Array.isArray(m.sourceCommits) && m.sourceCommits.length
+			? ` · <span class="commits">${escHtml(m.sourceCommits.map((c) => String(c).slice(0, 7)).join(", "))}</span>`
+			: "";
+	const showExisting =
+		m.existingBody !== null &&
+		m.existingBody !== undefined &&
+		["update", "delete", "keep"].includes(m.action);
+	return `<article class="card" data-id="${escHtml(m.id)}">
   <div class="card-head">
     <span class="card-title">${escHtml(m.title || m.id)}</span>
-    <span class="chip">${escHtml(m.type || '')}</span>
+    <span class="chip">${escHtml(m.type || "")}</span>
     <span class="badge ${cls}">${label}</span>
-    ${m.stale ? '<span class="badge stale">STALE</span>' : ''}
-    ${m.conflict ? `<button type="button" class="badge conflict conflict-btn" data-with="${escHtml(m.conflict.with || '')}">CONFLICT ⇄</button>` : ''}
+    ${m.stale ? '<span class="badge stale">STALE</span>' : ""}
+    ${m.conflict ? `<button type="button" class="badge conflict conflict-btn" data-with="${escHtml(m.conflict.with || "")}">CONFLICT ⇄</button>` : ""}
   </div>
   <div class="meta"><code>${escHtml(m.id)}.md</code>${tags}${commits}</div>
-  ${m.description ? `<div class="desc">${escHtml(m.description)}</div>` : ''}
-  ${m.stale && Array.isArray(m.staleReasons) && m.staleReasons.length
-    ? `<div class="stale-panel">${m.staleReasons.map(r => `<div>⚠ ${escHtml(r)}</div>`).join('')}</div>` : ''}
-  ${m.conflict
-    ? `<div class="conflict-panel" hidden>Conflicts with <code>${escHtml(m.conflict.with || '')}</code> — ${escHtml(m.conflict.summary || '')}</div>` : ''}
+  ${m.description ? `<div class="desc">${escHtml(m.description)}</div>` : ""}
+  ${m.reason ? `<div class="request-context"><strong>Why this change:</strong> ${escHtml(m.reason)}</div>` : ""}
+  ${
+		m.stale && Array.isArray(m.staleReasons) && m.staleReasons.length
+			? `<div class="stale-panel">${m.staleReasons.map((r) => `<div>⚠ ${escHtml(r)}</div>`).join("")}</div>`
+			: ""
+	}
+  ${
+		m.conflict
+			? `<div class="conflict-panel" hidden>Conflicts with <code>${escHtml(m.conflict.with || "")}</code> — ${escHtml(m.conflict.summary || "")}</div>`
+			: ""
+	}
   <div class="body md" data-body-id="${escHtml(m.id)}"></div>
-  ${showExisting
-    ? `<details class="existing"><summary>Current version on disk</summary><div class="md" data-existing-id="${escHtml(m.id)}"></div></details>` : ''}
+  ${
+		showExisting
+			? `<details class="existing"><summary>Current version on disk</summary><div class="md" data-existing-id="${escHtml(m.id)}"></div></details>`
+			: ""
+	}
   <div class="controls">
     <div class="verdicts" data-id="${escHtml(m.id)}">${verdictButtons(m)}</div>
     <textarea class="comment" data-id="${escHtml(m.id)}" rows="1"
@@ -99,27 +123,27 @@ function cardHtml(m) {
 }
 
 function areaSections(data, memories) {
-  const areas = Array.isArray(data.areas) ? data.areas : [];
-  const byArea = new Map(areas.map(a => [a.id, []]));
-  const other = [];
-  for (const m of memories) {
-    (byArea.get(m.area) || other).push(m);
-  }
-  let html = '';
-  for (const a of areas) {
-    const cards = byArea.get(a.id);
-    if (!cards.length && data.mode !== 'init') continue;
-    html += `<section class="area">
+	const areas = Array.isArray(data.areas) ? data.areas : [];
+	const byArea = new Map(areas.map((a) => [a.id, []]));
+	const other = [];
+	for (const m of memories) {
+		(byArea.get(m.area) || other).push(m);
+	}
+	let html = "";
+	for (const a of areas) {
+		const cards = byArea.get(a.id);
+		if (!cards.length && data.mode !== "init") continue;
+		html += `<section class="area">
 <h2>${escHtml(a.title || a.id)} <span class="count">${cards.length}</span></h2>
-${a.description ? `<p class="area-desc">${escHtml(a.description)}</p>` : ''}
-${cards.length ? cards.map(cardHtml).join('\n') : '<p class="empty">No memories drafted for this area.</p>'}
+${a.description ? `<p class="area-desc">${escHtml(a.description)}</p>` : ""}
+${cards.length ? cards.map(cardHtml).join("\n") : '<p class="empty">No memories drafted for this area.</p>'}
 </section>`;
-  }
-  if (other.length) {
-    html += `<section class="area"><h2>Other <span class="count">${other.length}</span></h2>
-${other.map(cardHtml).join('\n')}</section>`;
-  }
-  return html;
+	}
+	if (other.length) {
+		html += `<section class="area"><h2>Other <span class="count">${other.length}</span></h2>
+${other.map(cardHtml).join("\n")}</section>`;
+	}
+	return html;
 }
 
 const CSS = `
@@ -149,9 +173,15 @@ const CSS = `
 .meta code{font-size:12px}
 .tag{border:1px solid var(--border);border-radius:8px;padding:0 6px;font-size:11px}
 .desc{font-size:13px;color:var(--text-muted);font-style:italic;margin:2px 0 6px}
+.request-context{font-size:var(--fs-xs);line-height:1.45;background:var(--accent-soft);border-left:3px solid var(--accent);padding:6px 9px;margin:6px 0}.request-context strong{color:var(--accent)}
 .stale-panel{border-left:3px solid var(--dot-yellow);padding:4px 10px;margin:6px 0;font-size:13px;color:var(--dot-yellow)}
 .conflict-panel{border-left:3px solid var(--dot-red);padding:6px 10px;margin:6px 0;font-size:13px;color:var(--dot-red)}
 .body.md{border-top:1px solid var(--border);margin-top:8px;padding-top:4px;font-size:14px;overflow:hidden}
+.change-block{position:relative;margin:8px 0;padding:8px 10px;border-left:3px solid transparent;border-radius:0 var(--radius-sm) var(--radius-sm) 0}
+.change-block.added,.change-block.modified{background:var(--add-bg);border-left-color:var(--dot-green)}
+.change-block.removed{background:var(--del-bg);border-left-color:var(--dot-red);color:var(--del-fg)}
+.change-block.unchanged{padding-block:2px;opacity:.72}.change-label{font:600 10px var(--font-mono);letter-spacing:.05em;text-transform:uppercase;color:var(--text-muted);margin-bottom:4px}
+.change-block.added .change-label,.change-block.modified .change-label{color:var(--add-fg)}.change-block.removed .change-label{color:var(--del-fg)}
 .body.md.collapsed{max-height:280px;-webkit-mask-image:linear-gradient(black 75%,transparent);mask-image:linear-gradient(black 75%,transparent)}
 .show-more{display:block;margin:4px 0;font-size:12px;padding:2px 10px}
 details.existing{margin:8px 0;font-size:14px}
@@ -171,6 +201,7 @@ footer.general{margin:30px 0 60px}
 footer.general label{display:block;font-weight:600;margin-bottom:6px}
 #general-comment{width:100%;min-height:70px}
 .hint{color:var(--text-muted);font-size:12px;margin-top:6px}
+@media(max-width:640px){.change-block{margin-inline:0;padding:7px 8px}.controls{flex-direction:column}.verdicts{max-width:100%}}
 `;
 
 // Step client JS. No backticks / dollar-brace; data only via D.
@@ -206,7 +237,10 @@ function onReady() {
     var el = document.querySelector('[data-body-id="' + CSS.escape(m.id) + '"]');
     if (el) {
       var main = (m.action === 'keep' || m.body == null) ? (m.existingBody != null ? m.existingBody : m.body) : m.body;
-      el.innerHTML = mdToHtml(main == null ? '' : main);
+      var focusChanges = (m.action === 'create' || m.action === 'update') && m.body != null;
+      el.innerHTML = focusChanges
+        ? markdownChangeHtml(m.body, m.existingBody == null ? '' : m.existingBody, m.action === 'create')
+        : mdToHtml(main == null ? '' : main);
     }
     var ex = document.querySelector('[data-existing-id="' + CSS.escape(m.id) + '"]');
     if (ex) ex.innerHTML = mdToHtml(m.existingBody == null ? '' : m.existingBody);
@@ -257,16 +291,20 @@ onReady();
 `;
 
 export function render(data) {
-  const mode = ['init', 'consolidate', 'memorize'].includes(data.mode) ? data.mode : 'init';
-  const memories = Array.isArray(data.memories) ? data.memories : [];
-  const round = data.round || 1;
+	const mode = ["init", "consolidate", "memorize"].includes(data.mode)
+		? data.mode
+		: "init";
+	const memories = Array.isArray(data.memories) ? data.memories : [];
+	const round = data.round || 1;
 
-  const subtitleParts = [`/ ${mode} review`, `round ${round}`];
-  if (mode === 'memorize' && data.baseCommit && data.headCommit) {
-    subtitleParts.push(`${String(data.baseCommit).slice(0, 7)}..${String(data.headCommit).slice(0, 7)}`);
-  }
+	const subtitleParts = [`/ ${mode} review`, `round ${round}`];
+	if (mode === "memorize" && data.baseCommit && data.headCommit) {
+		subtitleParts.push(
+			`${String(data.baseCommit).slice(0, 7)}..${String(data.headCommit).slice(0, 7)}`,
+		);
+	}
 
-  const BODY = `<div class="wrap">
+	const BODY = `<div class="wrap">
 <div class="banner">${banner({ ...data, mode }, memories)}</div>
 <div id="areas">
 ${areaSections({ ...data, mode }, memories)}
@@ -280,16 +318,20 @@ ${areaSections({ ...data, mode }, memories)}
 </footer>
 </div>`;
 
-  return renderPage({
-    step: 'memory-review',
-    subtitle: subtitleParts.join(' · '),
-    branch: data.branch,
-    title: `${mode} review`,
-    data: { mode, memories, areas: Array.isArray(data.areas) ? data.areas : [] },
-    css: CSS,
-    body: BODY,
-    clientJs: JS,
-    primaryIdle: 'Accept',
-    primaryChanged: 'Send review',
-  });
+	return renderPage({
+		step: "memory-review",
+		subtitle: subtitleParts.join(" · "),
+		branch: data.branch,
+		title: `${mode} review`,
+		data: {
+			mode,
+			memories,
+			areas: Array.isArray(data.areas) ? data.areas : [],
+		},
+		css: CSS,
+		body: BODY,
+		clientJs: JS,
+		primaryIdle: "Accept",
+		primaryChanged: "Send review",
+	});
 }
diff --git a/skills/iterator/lib/views/plan.mjs b/skills/iterator/lib/views/plan.mjs
index 6b6f684..470bb12 100644
--- a/skills/iterator/lib/views/plan.mjs
+++ b/skills/iterator/lib/views/plan.mjs
@@ -11,12 +11,14 @@
  *     { type:"plan-approved"|"plan-feedback", sections, dependencies, comments, comment }
  *     plus the shared { type:"cancel" } / { type:"timeout" }.
  */
-import { renderPage } from '../ui.mjs';
+import { renderPage } from "../ui.mjs";
 
 const PLAN_CSS = `
 .main{max-width:960px;margin:0 auto;padding:28px var(--sp-5)}
 h1{font-family:var(--font-display);font-size:var(--fs-xl);font-weight:600;margin-bottom:6px}
 .hint{font-size:var(--fs-sm);color:var(--text-muted);margin-bottom:28px;line-height:1.5}
+.fast-track{display:inline-flex;margin:0 0 var(--sp-3);padding:3px 9px;border-radius:12px;background:var(--bg-green);color:var(--dot-green);font:600 var(--fs-xs) var(--font-mono)}
+body.fast-plan .cmt-btn,body.fast-plan .comment-section{display:none}
 #sections{display:grid;grid-template-columns:1fr 1fr;gap:var(--sp-5)}
 @media(max-width:720px){#sections{grid-template-columns:1fr}}
 .section{margin-bottom:var(--sp-4)}
@@ -82,7 +84,8 @@ textarea.comment-box:hover:not(:focus){border-color:var(--text-muted)}
 const PLAN_BODY = `
 <div class="main">
   <h1 id="title"></h1>
-  <p class="hint">Each section is rendered markdown — <strong>click any section to edit</strong> (save with blur or ⌘/Ctrl+Enter). Use the 💬 icon to leave a comment on a section, and confirm the <strong>external dependencies</strong> below — new packages, libraries, or services this plan needs, never tasks or todos. When everything looks right click <strong>Accept</strong>; if you edit anything or add a comment the button becomes <strong>Send review</strong> so Claude can revise first.</p>
+  <div class="fast-track" id="fast-track" hidden>Structured plan · ready for review</div>
+  <p class="hint" id="plan-hint">Each section is rendered markdown — <strong>click any section to edit</strong> (save with blur or ⌘/Ctrl+Enter). Use the 💬 icon to leave a comment on a section, and confirm the <strong>external dependencies</strong> below — new packages, libraries, or services this plan needs, never tasks or todos. When everything looks right click <strong>Accept</strong>; if you edit anything or add a comment the button becomes <strong>Send review</strong> so Claude can revise first.</p>
   <div class="deps-panel">
     <div class="shead"><span class="slabel">External dependencies</span><span class="chip-design" id="design-chip"></span></div>
     <div class="dep-chips" id="dep-chips"></div>
@@ -123,6 +126,11 @@ SECTION_KEYS.forEach(k => comments[k] = []);
 let activeKey = null;
 
 document.getElementById('title').textContent = D.title || 'Plan Review';
+if(D.fastTrack){
+  document.body.classList.add('fast-plan');
+  document.getElementById('fast-track').hidden = false;
+  document.getElementById('plan-hint').innerHTML = 'This complete plan skipped discovery and is ready for review. <strong>Click any section to edit it directly</strong>, confirm the external dependencies, then Accept to save it and continue to feature breakdown.';
+}
 const dchip = document.getElementById('design-chip');
 if(D.designFile){ dchip.className = 'chip-design on'; dchip.textContent = 'design params attached'; dchip.title = D.designFile; }
 else { dchip.className = 'chip-design off'; dchip.textContent = 'no design params \\u2014 /iterator-design'; dchip.title = 'UI-touching plans should capture memory/design.md first'; }
@@ -212,6 +220,7 @@ function deleteComment(i){ comments[activeKey].splice(i,1); renderComments(); re
 function collectComments(){ const out=[]; SECTION_KEYS.forEach(k => comments[k].forEach(text => out.push({section:k, text}))); return out; }
 
 function hasChanges(){
+  if(D.fastTrack) return false;
   if(document.getElementById('global-comment').value.trim()) return true;
   if(collectComments().length) return true;
   if(JSON.stringify(sections) !== ORIG_SECTIONS) return true;
@@ -232,9 +241,16 @@ function onPrimary(){
 `;
 
 export function render(data) {
-  return renderPage({
-    step: 'plan', subtitle: '/ plan review', branch: data.branch, title: data.title,
-    data, css: PLAN_CSS, body: PLAN_BODY, clientJs: PLAN_JS,
-    primaryIdle: 'Accept', primaryChanged: 'Send review',
-  });
+	return renderPage({
+		step: "plan",
+		subtitle: "/ plan review",
+		branch: data.branch,
+		title: data.title,
+		data,
+		css: PLAN_CSS,
+		body: PLAN_BODY,
+		clientJs: PLAN_JS,
+		primaryIdle: "Accept",
+		primaryChanged: "Send review",
+	});
 }
diff --git a/skills/iterator/lib/views/planning.mjs b/skills/iterator/lib/views/planning.mjs
index 7b9c314..6fc8f59 100644
--- a/skills/iterator/lib/views/planning.mjs
+++ b/skills/iterator/lib/views/planning.mjs
@@ -36,9 +36,11 @@ const PLANNING_CSS = `
 .backlog{background:var(--surface);border:1px solid var(--border);border-radius:var(--radius-card);box-shadow:var(--shadow-card);padding:var(--sp-4);margin-top:var(--sp-6)}
 .backlog-head{display:flex;align-items:center;justify-content:space-between;gap:var(--sp-3);flex-wrap:wrap;margin-bottom:var(--sp-3)}
 .backlog-head h2{font-size:var(--fs-md);font-weight:600}.backlog-head p{font-size:var(--fs-xs);color:var(--text-muted)}
+.backlog-tools{display:flex;align-items:center;gap:var(--sp-2);flex-wrap:wrap}.backlog-tools .filters{display:flex;gap:var(--sp-1)}
+.backlog-tools button.sel{background:var(--accent);border-color:var(--accent);color:var(--accent-fg)}
 .backlog-form{display:grid;grid-template-columns:120px minmax(180px,1fr);gap:var(--sp-2);align-items:start}
 .backlog-form select,.backlog-form input,.backlog-form textarea{width:100%;background:var(--bg);border:1px solid var(--border);border-radius:var(--radius-sm);padding:8px 10px;color:var(--text);font:inherit;font-size:var(--fs-sm)}
-.backlog-form textarea{min-height:64px;resize:vertical;grid-column:1/-1}.backlog-form .btns{grid-column:1/-1}.backlog-list{display:grid;gap:var(--sp-2);margin-top:var(--sp-4)}
+.backlog-form textarea{min-height:64px;resize:vertical;grid-column:1/-1}.backlog-form .at-wrap{position:relative;grid-column:1/-1}.backlog-form .at-wrap textarea{grid-column:auto}.backlog-form .btns{grid-column:1/-1}.backlog-list{display:grid;gap:var(--sp-2);margin-top:var(--sp-4)}
 .backlog-item{border-top:1px solid var(--border);padding-top:var(--sp-3)}.backlog-item:first-child{border-top:0;padding-top:0}
 .backlog-item-main{display:flex;gap:var(--sp-2);align-items:flex-start}.backlog-item-main input{margin-top:4px}.backlog-item-title{font-size:var(--fs-sm);font-weight:600}.backlog-item-details{font-size:var(--fs-xs);color:var(--text-muted);margin-top:2px;white-space:pre-wrap}.backlog-empty{font-size:var(--fs-sm);color:var(--text-muted)}
 .retired{margin-top:var(--sp-6)}.retired .sec-title{margin-top:0}.retired-list{display:grid;gap:var(--sp-3)}
@@ -56,11 +58,12 @@ function clearPlanDraft(){ try { localStorage.removeItem(PLAN_DRAFT_KEY); } catc
 
 async function backlogAction(payload, button, message){
   if(button){ button.disabled = true; button.dataset.label = button.textContent; button.textContent = 'Saving…'; }
-  await post({ type:'backlog', ...payload }, message || 'Saved', { allowWhileWorking:true });
+  const saved = await post({ type:'backlog', ...payload }, message || 'Saved', { allowWhileWorking:true });
   // During an agent turn the server deliberately does not refresh this view:
   // doing so would clear the model-working guard. Restore the local control so
   // more filesystem-only backlog edits remain possible until turn-end refresh.
   if(button){ button.disabled = false; if(button.dataset.label) button.textContent = button.dataset.label; delete button.dataset.label; }
+  return saved;
 }
 function selectedBacklogGoal(){
   const selected = (D.backlog || []).filter(item => item.selected);
@@ -68,6 +71,43 @@ function selectedBacklogGoal(){
   return 'Create a plan from these saved backlog candidates:\\n\\n' + selected.map(item =>
     '[' + item.kind + '] ' + item.title + (item.details ? '\\n' + item.details : '')).join('\\n\\n');
 }
+function parseStructuredPlan(value){
+  const lines = String(value || '').replaceAll('\\r', '').split('\\n');
+  const sections = {goal:[], architecture:[], keyDecisions:[], dependencies:[]};
+  let title = '', active = null, found = 0;
+  const heading = line => {
+    let text = line.trim();
+    const markdown = text.startsWith('#');
+    while(text.startsWith('#')) text = text.slice(1).trim();
+    if(text.endsWith(':')) text = text.slice(0, -1).trim();
+    const compact = text.toLowerCase().replaceAll(' ', '').replaceAll('-', '');
+    const keys = {goal:'goal',architecture:'architecture',keydecisions:'keyDecisions',decisions:'keyDecisions',dependencies:'dependencies',externaldependencies:'dependencies'};
+    return (markdown || line.trim().endsWith(':')) ? (keys[compact] || null) : null;
+  };
+  for(const raw of lines){
+    const trimmed = raw.trim();
+    const key = heading(raw);
+    if(key){ active = key; found++; continue; }
+    if(!title && trimmed.toLowerCase().startsWith('title:')){
+      title = trimmed.slice(6).trim(); continue;
+    }
+    if(!title && trimmed.startsWith('#') && !active){
+      title = trimmed.slice(1).trim(); continue;
+    }
+    if(active) sections[active].push(raw);
+  }
+  const clean = key => sections[key].join('\\n').trim();
+  const goal = clean('goal'), architecture = clean('architecture'), keyDecisions = clean('keyDecisions');
+  if(found < 3 || !goal || !architecture || !keyDecisions) return null;
+  if(!title) title = goal.split('\\n').find(Boolean).slice(0, 80);
+  const dependencies = sections.dependencies.map(line => {
+    let item = line.trim();
+    if(item.startsWith('- ') || item.startsWith('* ')) item = item.slice(2).trim();
+    return item;
+  }).filter(item => item && !['none','n/a','no new dependencies'].includes(item.toLowerCase()));
+  const description = goal.split('\\n').map(line => line.trim()).find(Boolean).slice(0, 160);
+  return {title, description, plan:{goal, architecture, keyDecisions}, dependencies};
+}
 
 // Minimal @-file suggestions on the goal box: substring filter over the
 // gathered tracked-file list, arrow keys + Enter/Tab to accept.
@@ -136,7 +176,7 @@ function render(){
       '<h2>No plan yet</h2><p>iterator keeps its state in a memory/ bundle in your repo.<br>Type a goal and start turning it into a reviewable plan.</p>';
     const goal = document.createElement('textarea');
     goal.className = 'goal';
-    goal.placeholder = 'What are you building and why? (1\\u20133 sentences \\u2014 optional, saves a question round; @ mentions repo files)';
+    goal.placeholder = 'Describe a goal, or paste a complete plan with Goal, Architecture, and Key Decisions headings. @ mentions repo files.';
     // The iframe is recreated on every tab switch/refresh — keep the unsent
     // goal in browser storage, cleared only once a plan actually starts.
     try { goal.value = localStorage.getItem(PLAN_DRAFT_KEY) || ''; } catch(e){}
@@ -172,9 +212,18 @@ function render(){
     }
     const b = document.createElement('button');
     b.className = 'act' + (D.knowledgeInitialized ? ' primary-act' : ''); b.textContent = 'Create plan';
-    b.addEventListener('click', () =>
-      post({ type:'action', action:'plan', feature:null, prompt: goal.value.trim() || null }, 'Starting /iterator-plan')
-        .then(() => { if(__submitted) clearPlanDraft(); }));
+    b.addEventListener('click', () => {
+      const prompt = goal.value.trim() || null;
+      const structuredPlan = parseStructuredPlan(prompt);
+      return post({
+        type:'action',
+        action:structuredPlan ? 'plan-fast-track' : 'plan',
+        feature:null,
+        prompt,
+        structuredPlan
+      }, structuredPlan ? 'Opening plan review' : 'Starting /iterator-plan')
+        .then(() => { if(__submitted && !structuredPlan) clearPlanDraft(); });
+    });
     btns.appendChild(b);
     hero.appendChild(btns);
     w.appendChild(hero);
@@ -183,65 +232,17 @@ function render(){
     return;
   }
 
-  // Plan bar: title, status, progress count, and the stage-driven lifecycle
-  // buttons. Execution controls (auto mode, Test/Implement/Review) live on
-  // the Work surface.
-  const done = (D.progress&&D.progress.done)||0, total = (D.progress&&D.progress.total)||CH.length;
-  const bar = document.createElement('div');
-  bar.className = 'planbar';
-  bar.innerHTML = '<span class="pt">'+esc(D.plan.title||'Plan')+'</span>'+
-    '<span class="pst '+(D.plan.status==='approved'?'approved':'draft')+'">'+esc(D.plan.status||'draft')+'</span>'+
-    (total
-      ? '<span class="pcount">'+done+' / '+total+' features done</span>'
-      : '<span class="pcount">not broken into features yet</span>');
-  const revise = document.createElement('button');
-  revise.className='act'; revise.textContent='Revise plan';
-  revise.addEventListener('click', () => action('plan', null, 'Starting /iterator-plan'));
-  bar.appendChild(revise);
-  // Before featuring, this button IS the continue action — make it read and
-  // look like one instead of a "Re-feature" that has nothing to redo.
-  const refeature = document.createElement('button');
-  refeature.className = D.stage==='needs-features' ? 'act primary-act' : 'act';
-  refeature.textContent = D.stage==='needs-features' ? 'Feature the plan' : 'Re-feature';
-  refeature.title = D.stage==='needs-features'
-    ? 'Next step: break the approved plan into small, dependency-ordered features (/iterator-feature)'
-    : 'Redraw the feature set from the plan (/iterator-feature)';
-  refeature.addEventListener('click', () => action('feature', null, 'Starting /iterator-feature'));
-  bar.appendChild(refeature);
-  // Everything landed → the whole-plan review; every feature done → retire.
-  // Both derive from the server-computed stage, never recomputed here.
-  if(D.stage==='awaiting-plan-review' || D.stage==='retirable'){
-    const rp = document.createElement('button');
-    rp.className = 'act primary-act';
-    rp.textContent = D.plan.planReviewed ? 'Re-review plan' : 'Review plan';
-    rp.title = D.plan.planReviewed
-      ? 'Plan reviewed '+D.plan.planReviewed+' \\u2014 run the whole-plan review again'
-      : 'Review all changes and commits against the plan\\u2019s goals and decisions';
-    rp.addEventListener('click', () => action('review-plan', null, 'Starting /iterator-review-plan'));
-    bar.appendChild(rp);
-  }
-  if(D.stage==='retirable'){
-    const retire = document.createElement('button');
-    retire.className='act primary-act'; retire.textContent='Retire plan';
-    retire.title='Condense the finished plan into a decisions/ memory and archive its features';
-    confirmButton(retire, 'Retires the plan \\u2014 click again', () =>
-      action('retire', null, 'Starting plan retirement'));
-    bar.appendChild(retire);
-  }
-  // Cancel: abandon the plan entirely — archives the bundle side and deletes
-  // the plan branch/worktree, so the armed label spells out what is at stake.
-  const cancelPlan = document.createElement('button');
-  cancelPlan.className = 'act danger';
-  cancelPlan.textContent = 'Cancel plan';
-  const dirtyWarn = (D.dirty && D.dirty.count)
-    ? D.dirty.count + ' uncommitted file' + (D.dirty.count!==1?'s':'') + ' + '
-    : '';
-  cancelPlan.title = 'Abandon this plan: archive it and DELETE its branch/worktree'
-    + (dirtyWarn ? ' \\u2014 \\u26a0 ' + dirtyWarn + 'unmerged commits will be lost' : '');
-  confirmButton(cancelPlan, '\\u26a0 Deletes ' + dirtyWarn + 'branch \\u2014 click again', () =>
-    action('cancel-plan', null, 'Cancelling plan'));
-  bar.appendChild(cancelPlan);
-  w.appendChild(bar);
+  // Planning is a staging surface once a plan exists. Active lifecycle,
+  // progress, and feature controls stay together on Work.
+  const hero = document.createElement('div');
+  hero.className = 'hero';
+  hero.innerHTML = '<h2>Planning is staged</h2><p><strong>'+esc(D.plan.title||'This plan')+'</strong> is active on Work.<br>Use Work for progress, feature actions, and plan lifecycle controls.</p>';
+  const open = document.createElement('button');
+  open.className = 'act primary-act';
+  open.textContent = 'Open Work';
+  open.addEventListener('click', () => action('hub', null, 'Opening Work'));
+  hero.appendChild(open);
+  w.appendChild(hero);
   renderBacklog(w);
   renderRetired(w);
 }
@@ -250,13 +251,15 @@ function render(){
 function renderBacklog(w){
   const items = Array.isArray(D.backlog) ? D.backlog : [];
   const section = document.createElement('section'); section.className = 'backlog backlog-active';
-  section.innerHTML = '<div class="backlog-head"><div><h2>Idea backlog</h2><p>Saved ideas and bugs stay separate from active plan features.</p></div></div>';
+  section.innerHTML = '<div class="backlog-head"><div><h2>Idea backlog</h2><p>Saved ideas and bugs stay separate from active plan features.</p></div><div class="backlog-tools"><div class="filters" aria-label="Filter backlog"><button class="act sel" type="button" data-kind="all">All</button><button class="act" type="button" data-kind="idea">Ideas</button><button class="act" type="button" data-kind="bug">Bugs</button></div><button class="act" type="button" data-bulk="select">Select visible</button><button class="act" type="button" data-bulk="deselect">Deselect visible</button></div></div>';
   const form = document.createElement('form'); form.className = 'backlog-form';
   form.innerHTML = '<select aria-label="Candidate type"><option value="idea">Idea</option><option value="bug">Bug</option></select>'+
     '<input required maxlength="160" placeholder="Short title" aria-label="Backlog title">'+
-    '<textarea maxlength="4000" placeholder="Why it matters, context, or a repro" aria-label="Backlog details"></textarea>'+
+    '<div class="at-wrap"><textarea maxlength="4000" placeholder="Why it matters, context, or a repro" aria-label="Backlog details"></textarea><div class="at-menu"></div></div>'+
     '<div class="btns"><button class="act primary-act" type="submit">Save candidate</button><button class="act" type="button" hidden>Cancel edit</button></div>';
   const [kind, title, details] = form.querySelectorAll('select,input,textarea');
+  const detailsMenu = form.querySelector('.at-menu');
+  wireAtMenu(details, detailsMenu);
   const [save, cancel] = form.querySelectorAll('button');
   let editing = null;
   const reset = () => { editing = null; form.reset(); save.textContent = 'Save candidate'; cancel.hidden = true; };
@@ -267,12 +270,21 @@ function renderBacklog(w){
   cancel.addEventListener('click', reset);
   section.appendChild(form);
   const list = document.createElement('div'); list.className = 'backlog-list bounded-list';
+  const rows = new Map();
+  let filter = 'all';
+  let refreshHandoff = () => {};
   if(!items.length) list.innerHTML = '<p class="backlog-empty">No saved candidates yet.</p>';
   for(const item of items){
-    const row = document.createElement('div'); row.className = 'backlog-item';
+    const row = document.createElement('div'); row.className = 'backlog-item'; row.dataset.kind = item.kind;
+    rows.set(item.id, row);
     row.innerHTML = '<div class="backlog-item-main"><input type="checkbox" aria-label="Select '+esc(item.title)+'" '+(item.selected?'checked':'')+'><div><div class="backlog-item-title">'+esc(item.title)+' <span class="chip cmut">'+esc(item.kind)+'</span></div><div class="backlog-item-details">'+esc(item.details || '')+'</div></div></div>';
     const toggle = row.querySelector('input');
-    toggle.addEventListener('change', () => backlogAction({ action:'select', id:item.id, selected:toggle.checked }, toggle, toggle.checked ? 'Candidate selected' : 'Candidate deselected'));
+    toggle.addEventListener('change', async () => {
+      const selected = toggle.checked;
+      const saved = await backlogAction({ action:'select', id:item.id, selected }, toggle, selected ? 'Candidate selected' : 'Candidate deselected');
+      if(saved){ item.selected = selected; refreshHandoff(); }
+      else toggle.checked = !selected;
+    });
     const buttons = document.createElement('div'); buttons.className = 'btns';
     const edit = document.createElement('button'); edit.className = 'act'; edit.type = 'button'; edit.textContent = 'Edit';
     edit.addEventListener('click', () => { editing = item.id; kind.value = item.kind; title.value = item.title; details.value = item.details || ''; save.textContent = 'Update candidate'; cancel.hidden = false; title.focus(); });
@@ -280,16 +292,43 @@ function renderBacklog(w){
     confirmButton(remove, 'Really delete — click again', () => backlogAction({ action:'delete', id:item.id }, remove, 'Candidate deleted'));
     buttons.append(edit, remove); row.appendChild(buttons); list.appendChild(row);
   }
+  const visibleItems = () => items.filter(item => filter==='all' || item.kind===filter);
+  const applyFilter = kind => {
+    filter = kind;
+    section.querySelectorAll('.filters [data-kind]').forEach(button => button.classList.toggle('sel', button.dataset.kind===kind));
+    rows.forEach((row, id) => { row.hidden = !visibleItems().some(item => item.id===id); });
+  };
+  section.querySelectorAll('.filters [data-kind]').forEach(button =>
+    button.addEventListener('click', () => applyFilter(button.dataset.kind)));
+  const bulkSelect = async (selected, button) => {
+    const targets = visibleItems().filter(item => Boolean(item.selected)!==selected);
+    if(!targets.length) return;
+    const saved = await backlogAction(
+      { action:'select-many', ids:targets.map(item => item.id), selected },
+      button,
+      selected ? 'Visible candidates selected' : 'Visible candidates deselected'
+    );
+    if(!saved) return;
+    for(const item of targets){
+      item.selected = selected;
+      const checkbox = rows.get(item.id).querySelector('input'); checkbox.checked = selected;
+    }
+    refreshHandoff();
+  };
+  section.querySelector('[data-bulk="select"]').addEventListener('click', event => bulkSelect(true, event.currentTarget));
+  section.querySelector('[data-bulk="deselect"]').addEventListener('click', event => bulkSelect(false, event.currentTarget));
   section.appendChild(list);
-  const goal = selectedBacklogGoal();
-  if(goal){
-    const handoff = document.createElement('button'); handoff.className = 'act primary-act'; handoff.type = 'button';
-    handoff.textContent = D.plan ? 'Selected candidates saved' : 'Plan selected candidates';
-    handoff.title = D.plan ? 'Retire or finish the active plan before starting a new one.' : 'Start a new plan with the selected candidates as its initial goal. They are removed from the backlog only after the plan is approved.';
-    handoff.disabled = Boolean(D.plan);
-    handoff.addEventListener('click', () => action('plan', null, 'Starting /iterator-plan from backlog', goal));
-    section.appendChild(handoff);
-  }
+  const handoff = document.createElement('button'); handoff.className = 'act primary-act'; handoff.type = 'button';
+  handoff.textContent = D.plan ? 'Selected candidates saved' : 'Plan selected candidates';
+  handoff.title = D.plan ? 'Retire or finish the active plan before starting a new one.' : 'Start a new plan with the selected candidates as its initial goal. They are removed from the backlog only after the plan is approved.';
+  handoff.disabled = Boolean(D.plan);
+  handoff.addEventListener('click', () => {
+    const goal = selectedBacklogGoal();
+    if(goal) action('plan', null, 'Starting /iterator-plan from backlog', goal);
+  });
+  refreshHandoff = () => { handoff.hidden = !selectedBacklogGoal(); };
+  refreshHandoff();
+  section.appendChild(handoff);
   w.appendChild(section);
 }
 
diff --git a/skills/iterator/lib/views/settings.mjs b/skills/iterator/lib/views/settings.mjs
index 2050b1c..8ae1d49 100644
--- a/skills/iterator/lib/views/settings.mjs
+++ b/skills/iterator/lib/views/settings.mjs
@@ -10,10 +10,10 @@
  *             models?: [{id,label}] }               // pi model registry, for model dropdowns
  *   output: one JSON line to stdout —
  *     { type:"settings", values:{<changed key>:<value>} }
- *     plus the shared { type:"cancel" } / { type:"timeout" }.
+ *     | { type:"settings-close" } when the shell-owned modal is dismissed.
  */
-import { renderPage } from '../ui.mjs';
-import { SETTINGS_DEFS } from '../settings.mjs';
+import { renderPage } from "../ui.mjs";
+import { SETTINGS_DEFS } from "../settings.mjs";
 
 const CSS = `
 .main{max-width:760px;margin:0 auto;padding:28px var(--sp-5)}
@@ -48,11 +48,41 @@ const BODY = `
 
 // Render order: behavior first, then models, then the knobs.
 const GROUPS = [
-  ['Flow', ['auto_mode', 'testing_default', 'review_required', 'branch_per_plan', 'worktree_per_plan', 'auto_retire_prompt']],
-  ['Models', ['planner_model', 'planner_thinking', 'implementer_model', 'implementer_thinking',
-    'tester_model', 'tester_thinking', 'reviewer_model', 'reviewer_thinking',
-    'plan_reviewer_model', 'plan_reviewer_thinking']],
-  ['Limits & bookkeeping', ['max_review_iterations', 'block_commit_on_leftovers', 'memorize_nudge', 'usage_ledger']],
+	[
+		"Flow",
+		[
+			"auto_mode",
+			"testing_default",
+			"review_required",
+			"branch_per_plan",
+			"worktree_per_plan",
+			"auto_retire_prompt",
+		],
+	],
+	[
+		"Models",
+		[
+			"planner_model",
+			"planner_thinking",
+			"implementer_model",
+			"implementer_thinking",
+			"tester_model",
+			"tester_thinking",
+			"reviewer_model",
+			"reviewer_thinking",
+			"plan_reviewer_model",
+			"plan_reviewer_thinking",
+		],
+	],
+	[
+		"Limits & bookkeeping",
+		[
+			"max_review_iterations",
+			"block_commit_on_leftovers",
+			"memorize_nudge",
+			"usage_ledger",
+		],
+	],
 ];
 
 const JS = `
@@ -153,15 +183,26 @@ function hasChanges(){
 }
 function onPrimary(){
   const values = changedValues();
-  if(!Object.keys(values).length){ post({ type:'cancel' }, 'Nothing changed'); return; }
+  if(!Object.keys(values).length){
+    post({ type:'settings-close' }, 'Settings closed', { allowWhileWorking:true });
+    return;
+  }
   post({ type:'settings', values }, 'Settings saved');
 }
 `;
 
 export function render(data) {
-  return renderPage({
-    step: 'settings', subtitle: '/ settings', branch: data.branch, title: data.plan,
-    data, css: CSS, body: BODY, clientJs: JS,
-    primaryIdle: 'Close', primaryChanged: 'Save settings',
-  });
+	return renderPage({
+		step: "settings",
+		subtitle: "/ settings",
+		branch: data.branch,
+		title: data.plan,
+		data,
+		css: CSS,
+		body: BODY,
+		clientJs: JS,
+		primaryIdle: "Close",
+		primaryChanged: "Save settings",
+		cancel: false,
+	});
 }
diff --git a/skills/iterator/lib/views/widgets.mjs b/skills/iterator/lib/views/widgets.mjs
index de8e9d7..18a6280 100644
--- a/skills/iterator/lib/views/widgets.mjs
+++ b/skills/iterator/lib/views/widgets.mjs
@@ -64,7 +64,7 @@ function action(act, feature, msg, prompt){
 function testBadge(c){
   if(!c.testsStatus || c.testsStatus==='none') return '';
   return c.testsStatus==='green' ? '<span class="chip cg"><i class="sdot"></i>tests green</span>'
-                                 : '<span class="chip cr"><i class="sdot"></i>tests red</span>';
+                                 : '<span class="chip cr"><i class="sdot"></i>tests committed · intentionally red</span>';
 }
 function sizeChip(c){
   const cls = c.size==='large'?'cr':c.size==='medium'?'cy':'cg';
diff --git a/skills/iterator/lib/write.mjs b/skills/iterator/lib/write.mjs
index 8f04e37..ee837e1 100644
--- a/skills/iterator/lib/write.mjs
+++ b/skills/iterator/lib/write.mjs
@@ -898,16 +898,28 @@ function writeBacklog(payload, root) {
 			updated: nowIso(),
 		};
 		items[i] = changed;
-	} else if (action === "select") {
-		const id = String(payload.id || "").trim();
-		if (!/^[a-z0-9][a-z0-9-]*$/.test(id))
+	} else if (action === "select" || action === "select-many") {
+		const ids =
+			action === "select"
+				? [String(payload.id || "").trim()]
+				: Array.isArray(payload.ids)
+					? [...new Set(payload.ids.map((id) => String(id).trim()))]
+					: [];
+		if (!ids.length) fail("backlog ids must be a non-empty array");
+		if (ids.some((id) => !/^[a-z0-9][a-z0-9-]*$/.test(id)))
 			fail("backlog id must be a kebab-case slug");
 		if (typeof payload.selected !== "boolean")
 			fail("backlog selected must be a boolean");
-		const i = items.findIndex((item) => item.id === id);
-		if (i === -1) fail(`no backlog item '${id}'`);
-		changed = { ...items[i], selected: payload.selected, updated: nowIso() };
-		items[i] = changed;
+		const missing = ids.filter((id) => !items.some((item) => item.id === id));
+		if (missing.length) fail(`no backlog item '${missing[0]}'`);
+		const changedIds = new Set(ids);
+		const updated = nowIso();
+		changed = [];
+		for (let i = 0; i < items.length; i++) {
+			if (!changedIds.has(items[i].id)) continue;
+			items[i] = { ...items[i], selected: payload.selected, updated };
+			changed.push(items[i]);
+		}
 	} else if (action === "delete") {
 		const id = String(payload.id || "").trim();
 		if (!/^[a-z0-9][a-z0-9-]*$/.test(id))
@@ -917,9 +929,10 @@ function writeBacklog(payload, root) {
 		changed = items[i];
 		items.splice(i, 1);
 	} else {
-		fail("backlog action must be create, edit, select, or delete");
+		fail("backlog action must be create, edit, select, select-many, or delete");
 	}
 
+	const changedItems = Array.isArray(changed) ? changed : [changed];
 	const dir = join(b.memDir, "backlog");
 	mkdirSync(dir, { recursive: true });
 	writeFileSync(backlogPath(b), backlogIndex(items));
@@ -927,9 +940,16 @@ function writeBacklog(payload, root) {
 	prependLog(
 		b.memDir,
 		payload.log ||
-			`**Backlog**: ${action} ${changed.id}${action === "select" ? (changed.selected ? " (selected)" : " (deselected)") : ""}.`,
+			`**Backlog**: ${action} ${changedItems.map((item) => item.id).join(", ")}${action.startsWith("select") ? (payload.selected ? " (selected)" : " (deselected)") : ""}.`,
 	);
-	return { op: "backlog", action, item: changed, items, memoryDir: b.memDir };
+	return {
+		op: "backlog",
+		action,
+		item: changedItems[0],
+		changedItems,
+		items,
+		memoryDir: b.memDir,
+	};
 }
 
 // ---------------------------------------------------------------------------
@@ -3207,8 +3227,9 @@ const SCHEMAS = {
 	},
 	backlog: {
 		op: "backlog",
-		action: "create|edit|select|delete",
-		"id?": "kebab-case item id (required except create)",
+		action: "create|edit|select|select-many|delete",
+		"id?": "kebab-case item id (required for edit/select/delete)",
+		"ids?": ["item ids (required for select-many)"],
 		"title?": "1-160 characters (required for create/edit)",
 		"details?": "up to 4000 characters (required for create/edit)",
 		"kind?": "idea|bug (required for create/edit)",
diff --git a/test/extension-backlog-save.test.mjs b/test/extension-backlog-save.test.mjs
new file mode 100644
index 0000000..0d11ca0
--- /dev/null
+++ b/test/extension-backlog-save.test.mjs
@@ -0,0 +1,27 @@
+import assert from "node:assert/strict";
+import { readFileSync } from "node:fs";
+import { test } from "node:test";
+
+const source = readFileSync(
+	new URL("../extensions/iterator.js", import.meta.url),
+	"utf8",
+);
+
+test("backlog saves refresh stored dashboards without clearing active work", () => {
+	const from = source.indexOf("const saveBacklog = async (input) => {");
+	const to = source.indexOf("/** Persist the Usage tab", from);
+	assert.notEqual(from, -1);
+	assert.notEqual(to, -1);
+	const section = source.slice(from, to);
+	assert.match(
+		section,
+		/const preserveAgentWorking = session\?\.isWorking\?\.\(\) === true/,
+	);
+	assert.match(section, /ids: input\.ids/);
+	assert.match(section, /result\.action === "select-many"/);
+	assert.match(section, /result\.changedItems\.length/);
+	assert.match(
+		section,
+		/if \(!preserveAgentWorking\) session\?\.clearWorking\?\.\(\);\s*await refreshHub\(cwd\);/,
+	);
+});
diff --git a/test/extension-conflict-resolution.test.mjs b/test/extension-conflict-resolution.test.mjs
new file mode 100644
index 0000000..a086031
--- /dev/null
+++ b/test/extension-conflict-resolution.test.mjs
@@ -0,0 +1,30 @@
+import assert from "node:assert/strict";
+import { readFileSync } from "node:fs";
+import { test } from "node:test";
+
+const source = readFileSync(
+	new URL("../extensions/iterator.js", import.meta.url),
+	"utf8",
+);
+
+test("Work conflicts dispatch reviewed targeted decision updates", () => {
+	const from = source.indexOf("const startConflictResolution = (result) => {");
+	const to = source.indexOf("/**\n\t * Apply a role", from);
+	assert.notEqual(from, -1);
+	assert.notEqual(to, -1);
+	const section = source.slice(from, to);
+	assert.match(section, /\/skill:iterator-knowledge update-memory/);
+	assert.match(section, /explicit memory verdict/);
+	assert.match(section, /Re-check against anchored files/);
+	assert.match(section, /re-check only \$\{feature\} against \$\{target\}/);
+	assert.match(section, /preserving every unrelated feature and conflict flag/);
+	assert.doesNotMatch(section, /iterator_write|runJson\(scriptPath\("write"\)/);
+});
+
+test("the dashboard intercepts conflict resolution before generic actions", () => {
+	const marker = 'result.action === "resolve-memory-conflict"';
+	const at = source.indexOf(marker);
+	assert.notEqual(at, -1);
+	const route = source.slice(at, at + 180);
+	assert.match(route, /startConflictResolution\(result\)/);
+});
diff --git a/test/extension-fresh-session.test.mjs b/test/extension-fresh-session.test.mjs
new file mode 100644
index 0000000..ea79c4a
--- /dev/null
+++ b/test/extension-fresh-session.test.mjs
@@ -0,0 +1,80 @@
+import { test } from "node:test";
+import assert from "node:assert/strict";
+import { createRequire } from "node:module";
+
+const require = createRequire(import.meta.url);
+const extensionDependenciesAvailable = (() => {
+	try {
+		require.resolve("typebox");
+		return true;
+	} catch {
+		return false;
+	}
+})();
+
+async function loadExtension() {
+	const { default: register } = await import("../extensions/iterator.js");
+	return register;
+}
+
+function mockPi() {
+	const commands = new Map();
+	return {
+		pi: {
+			on() {},
+			registerTool() {},
+			registerCommand(name, definition) {
+				commands.set(name, definition);
+			},
+			sendUserMessage() {},
+			setThinkingLevel() {},
+			async setModel() {
+				return true;
+			},
+		},
+		commands,
+	};
+}
+
+test("iterator-implement starts a clean replacement session with only its feature command", {
+	skip: !extensionDependenciesAvailable,
+}, async () => {
+	const { pi, commands } = mockPi();
+	const register = await loadExtension();
+	register(pi);
+	const marker = [];
+	const sent = [];
+	const ctx = {
+		hasUI: true,
+		sessionManager: { getSessionFile: () => "/tmp/parent.jsonl" },
+		async newSession(options) {
+			assert.equal(options.parentSession, "/tmp/parent.jsonl");
+			await options.setup({
+				appendCustomEntry(type, data) {
+					marker.push({ type, data });
+				},
+			});
+			await options.withSession({
+				async sendUserMessage(command) {
+					sent.push(command);
+				},
+			});
+			return { cancelled: false };
+		},
+	};
+
+	await commands.get("iterator-implement").handler("auth --auto", ctx);
+
+	assert.deepEqual(sent, ["/skill:iterator-implement auth --auto"]);
+	assert.deepEqual(marker, [
+		{
+			type: "iterator-implementation-handoff",
+			data: {
+				feature: "auth",
+				auto: true,
+				autoSteps: 0,
+				featureWave: null,
+			},
+		},
+	]);
+});
diff --git a/test/extension-model-lifecycle.test.mjs b/test/extension-model-lifecycle.test.mjs
new file mode 100644
index 0000000..1b017b4
--- /dev/null
+++ b/test/extension-model-lifecycle.test.mjs
@@ -0,0 +1,145 @@
+import { test } from "node:test";
+import assert from "node:assert/strict";
+import { execFileSync } from "node:child_process";
+import { createRequire } from "node:module";
+import { mkdtempSync, mkdirSync, rmSync, writeFileSync } from "node:fs";
+import { tmpdir } from "node:os";
+import { join } from "node:path";
+
+const require = createRequire(import.meta.url);
+const extensionDependenciesAvailable = (() => {
+	try {
+		require.resolve("typebox");
+		return true;
+	} catch {
+		return false;
+	}
+})();
+
+async function iteratorExtension(pi) {
+	const { default: register } = await import("../extensions/iterator.js");
+	return register(pi);
+}
+
+function fixture(settings = {}) {
+	const root = mkdtempSync(join(tmpdir(), "iterator-role-model-"));
+	execFileSync("git", ["init", "--quiet"], { cwd: root });
+	mkdirSync(join(root, "memory"), { recursive: true });
+	writeFileSync(
+		join(root, "memory", "plan.md"),
+		"---\ntype: Plan\ntitle: Role model test\nstatus: approved\n---\n\n# Goal\nTest role models.\n",
+	);
+	writeFileSync(
+		join(root, "memory", "settings.md"),
+		`---\ntype: Settings\n${Object.entries(settings)
+			.map(([key, value]) => `${key}: ${value}`)
+			.join("\n")}\n---\n`,
+	);
+	writeFileSync(
+		join(root, "memory", "state.md"),
+		"---\ntype: State\nmode: manual\n---\n",
+	);
+	return root;
+}
+
+function mockPi() {
+	const handlers = new Map();
+	const setModels = [];
+	const currentModel = { provider: "proxy", id: "managed" };
+	const overrideModel = { provider: "openai", id: "override" };
+	const pi = {
+		on(name, handler) {
+			handlers.set(name, handler);
+		},
+		registerCommand() {},
+		registerTool() {},
+		setThinkingLevel() {},
+		async setModel(model) {
+			setModels.push(model);
+			return model !== overrideModel || pi.overrideSucceeds;
+		},
+	};
+	return { pi, handlers, setModels, currentModel, overrideModel };
+}
+
+async function startTurn(handlers, root, modelRegistry, model) {
+	await handlers.get("before_agent_start")(
+		{},
+		{
+			cwd: root,
+			hasUI: false,
+			modelRegistry,
+			model,
+		},
+	);
+}
+
+test("failed tester override does not restore or alter the following active implementer", {
+	skip: !extensionDependenciesAvailable,
+}, async () => {
+	const root = fixture({ tester_model: "openai/override" });
+	try {
+		const { pi, handlers, setModels, currentModel, overrideModel } = mockPi();
+		await iteratorExtension(pi);
+		const registry = { find: () => overrideModel };
+
+		await handlers.get("input")({
+			text: "/iterator-test safe-role-model-handoff",
+		});
+		await startTurn(handlers, root, registry, currentModel);
+		await handlers.get("agent_end")({}, { cwd: root, hasUI: false });
+
+		await handlers.get("input")({
+			text: "/iterator-implement safe-role-model-handoff",
+		});
+		await startTurn(handlers, root, registry, currentModel);
+		await handlers.get("agent_end")({}, { cwd: root, hasUI: false });
+
+		assert.deepEqual(setModels, [overrideModel]);
+	} finally {
+		rmSync(root, { recursive: true, force: true });
+	}
+});
+
+test("successful manual override restores once and a role input is consumed by one turn", {
+	skip: !extensionDependenciesAvailable,
+}, async () => {
+	const root = fixture({ tester_model: "openai/override" });
+	try {
+		const { pi, handlers, setModels, currentModel, overrideModel } = mockPi();
+		pi.overrideSucceeds = true;
+		await iteratorExtension(pi);
+		const registry = { find: () => overrideModel };
+
+		await handlers.get("input")({
+			text: "/iterator-test safe-role-model-handoff",
+		});
+		await startTurn(handlers, root, registry, currentModel);
+		await startTurn(handlers, root, registry, currentModel);
+		await handlers.get("agent_end")({}, { cwd: root, hasUI: false });
+
+		assert.deepEqual(setModels, [overrideModel, currentModel]);
+	} finally {
+		rmSync(root, { recursive: true, force: true });
+	}
+});
+
+test("active role settings never call setModel", {
+	skip: !extensionDependenciesAvailable,
+}, async () => {
+	const root = fixture({ implementer_model: "active" });
+	try {
+		const { pi, handlers, setModels, currentModel } = mockPi();
+		await iteratorExtension(pi);
+
+		await handlers.get("input")({
+			text: "/iterator-implement safe-role-model-handoff",
+		});
+		await startTurn(handlers, root, { find: () => null }, currentModel);
+		await handlers.get("agent_end")({}, { cwd: root, hasUI: false });
+
+		assert.deepEqual(setModels, []);
+	} finally {
+		rmSync(root, { recursive: true, force: true });
+	}
+});
diff --git a/test/extension-plan-fast-track.test.mjs b/test/extension-plan-fast-track.test.mjs
new file mode 100644
index 0000000..be80e96
--- /dev/null
+++ b/test/extension-plan-fast-track.test.mjs
@@ -0,0 +1,42 @@
+import assert from "node:assert/strict";
+import { readFileSync } from "node:fs";
+import { test } from "node:test";
+
+const source = readFileSync(
+	new URL("../extensions/iterator.js", import.meta.url),
+	"utf8",
+);
+
+test("structured plans keep review and writer gates before feature breakdown", () => {
+	const from = source.indexOf("const startFastPlan = async (input) => {");
+	const to = source.indexOf("/** Persist the Usage tab", from);
+	assert.notEqual(from, -1);
+	assert.notEqual(to, -1);
+	const section = source.slice(from, to);
+	assert.match(section, /session\.showStep\(\{/);
+	assert.match(section, /step: "plan"/);
+	assert.match(section, /result\?\.type !== "plan-approved"/);
+	assert.match(section, /op: "plan"/);
+	assert.match(section, /sections: result\.sections/);
+	assert.match(section, /dependencies: result\.dependencies \|\| \[\]/);
+	assert.match(section, /refreshHub\(cwd, \{ activateWork: true \}\)/);
+	assert.match(section, /dispatch\("\/skill:iterator-feature"\)/);
+	assert.ok(
+		section.indexOf('result?.type !== "plan-approved"') <
+			section.indexOf('op: "plan"'),
+		"approval precedes the deterministic write",
+	);
+	assert.ok(
+		section.indexOf('op: "plan"') <
+			section.indexOf('dispatch("/skill:iterator-feature")'),
+		"feature breakdown starts only after the approved write",
+	);
+});
+
+test("dashboard routes the fast-track action without starting a planner model", () => {
+	const route = source.slice(
+		source.indexOf('result.action === "plan-fast-track"') - 100,
+		source.indexOf('result.action === "plan-fast-track"') + 180,
+	);
+	assert.match(route, /void startFastPlan\(result\)/);
+});
diff --git a/test/extension-red-test-handoff.test.mjs b/test/extension-red-test-handoff.test.mjs
new file mode 100644
index 0000000..9d3f6ca
--- /dev/null
+++ b/test/extension-red-test-handoff.test.mjs
@@ -0,0 +1,21 @@
+import assert from "node:assert/strict";
+import { readFileSync } from "node:fs";
+import { test } from "node:test";
+
+const source = readFileSync(
+	new URL("../extensions/iterator.js", import.meta.url),
+	"utf8",
+);
+
+test("committed tests intentionally refresh and activate Work", () => {
+	const from = source.indexOf(
+		'if (params.op === "commit-tests" && result?.ok)',
+	);
+	const to = source.indexOf("if (approved)", from);
+	assert.notEqual(from, -1, "commit-tests handoff is registered");
+	assert.notEqual(to, -1, "test handoff remains separate from plan approval");
+	assert.match(
+		source.slice(from, to),
+		/refreshHub\(ctx\.cwd, \{ activateWork: true \}\)/,
+	);
+});
diff --git a/test/extension-work-activation.test.mjs b/test/extension-work-activation.test.mjs
new file mode 100644
index 0000000..0051ebd
--- /dev/null
+++ b/test/extension-work-activation.test.mjs
@@ -0,0 +1,40 @@
+import { readFileSync } from "node:fs";
+import { test } from "node:test";
+import assert from "node:assert/strict";
+
+const source = readFileSync(
+	new URL("../extensions/iterator.js", import.meta.url),
+	"utf8",
+);
+
+function sourceSection(start, end) {
+	const from = source.indexOf(start);
+	const to = source.indexOf(end, from + start.length);
+	assert.notEqual(from, -1, `missing section start: ${start}`);
+	assert.notEqual(to, -1, `missing section end: ${end}`);
+	return source.slice(from, to);
+}
+
+test("accepted feature breakdown intentionally activates Work", () => {
+	const section = sourceSection(
+		"// Issue 5: auto mode starts right after the feature set is approved",
+		'pi.registerTool({\n\t\tname: "okf_write"',
+	);
+	assert.match(section, /const approved\s*=/);
+	assert.match(
+		section,
+		/if \(approved\) \{\s*void refreshHub\(ctx\.cwd, \{ activateWork: true \}\);/,
+	);
+});
+
+test("approved plan application intentionally activates Work", () => {
+	const section = sourceSection(
+		"// Apply-on-approve (mirrors lib/app.mjs)",
+		"return asText(result);",
+	);
+	assert.match(section, /result\?\.type === "plan-approved"/);
+	assert.match(
+		section,
+		/if \(applied\?\.ok\) await refreshHub\(ctx\.cwd, \{ activateWork: true \}\);/,
+	);
+});
diff --git a/test/gather.test.mjs b/test/gather.test.mjs
index b05e6dc..8e286e4 100644
--- a/test/gather.test.mjs
+++ b/test/gather.test.mjs
@@ -162,6 +162,8 @@ test("gather builds the hub payload from bundle + git state", () => {
 		assert.equal(config.ready, true);
 
 		assert.equal(auth.status, "pending");
+		assert.deepEqual(auth.tests, []);
+		assert.equal(auth.testCount, 0);
 		assert.deepEqual(auth.dependsOn, ["config-module"]);
 		assert.equal(auth.ready, true, "done dependency satisfies");
 		assert.deepEqual(
@@ -411,6 +413,31 @@ test("globToRegExp handles exact paths, * and **", () => {
 	assert.ok(globToRegExp("src/**").test("src/deep/a.ts"));
 });
 
+test("hub exposes recorded red-test paths and count as implementation targets", () => {
+	const root = makeFixture();
+	try {
+		const feature = join(root, "memory", "features", "auth-middleware.md");
+		writeFileSync(
+			feature,
+			readFileSync(feature, "utf8").replace(
+				'files: ["src/auth/*.ts"]',
+				'tests_status: red\ntests: ["test/auth.test.mjs", "test/auth-policy.test.mjs"]\nfiles: ["src/auth/*.ts"]',
+			),
+		);
+		const auth = gather(root).features.find(
+			(feature) => feature.name === "auth-middleware",
+		);
+		assert.equal(auth.testsStatus, "red");
+		assert.deepEqual(auth.tests, [
+			"test/auth.test.mjs",
+			"test/auth-policy.test.mjs",
+		]);
+		assert.equal(auth.testCount, 2);
+	} finally {
+		rmSync(root, { recursive: true, force: true });
+	}
+});
+
 test("implement rounds carry exactly ONE feature (next) with its full contract", () => {
 	const root = makeFixture();
 	try {
@@ -1153,7 +1180,7 @@ test("feature payload surfaces decisions concepts and stored memories/conflicts"
 		mkdirSync(join(root, "memory", "decisions"), { recursive: true });
 		writeFileSync(
 			join(root, "memory", "decisions", "no-orm.md"),
-			"---\ntype: Decision\ntitle: No ORM\ndescription: Raw SQL only.\n---\n\nbody\n",
+			'---\ntype: Decision\ntitle: No ORM\ndescription: Raw SQL only.\nfiles: ["src/auth/*.ts"]\n---\n\nbody\n',
 		);
 		writeFileSync(
 			join(root, "memory", "features", "auth-middleware.md"),
@@ -1177,6 +1204,19 @@ conflicts: '[{"decision":"decisions/no-orm","note":"introduces an ORM"}]'
 		assert.deepEqual(auth.conflicts, [
 			{ decision: "decisions/no-orm", note: "introduces an ORM" },
 		]);
+		const hubAuth = gather(root).features.find(
+			(c) => c.name === "auth-middleware",
+		);
+		assert.deepEqual(hubAuth.conflicts, [
+			{
+				decision: "decisions/no-orm",
+				note: "introduces an ORM",
+				title: "No ORM",
+				description: "Raw SQL only.",
+				files: ["src/auth/*.ts"],
+				decisionExists: true,
+			},
+		]);
 
 		// Implement contract unions the stored list with the dynamic match.
 		const imp = gatherImplement(root);
diff --git a/test/pi-tools.test.mjs b/test/pi-tools.test.mjs
index fa931c4..6a0a05b 100644
--- a/test/pi-tools.test.mjs
+++ b/test/pi-tools.test.mjs
@@ -6,6 +6,9 @@ import { join } from "node:path";
 import {
 	actionToCommand,
 	activityTextFromMessage,
+	implementationCommand,
+	implementationHandoffState,
+	shouldApplyRole,
 	bundleExists,
 	featuresDirEntries,
 	composeAmbientContext,
@@ -43,7 +46,7 @@ test("actionToCommand maps hub actions to skill commands", () => {
 	);
 	assert.equal(
 		actionToCommand({ type: "action", action: "implement", feature: "auth" }),
-		"/skill:iterator-implement auth",
+		"/iterator-implement auth",
 	);
 	assert.equal(
 		actionToCommand({ type: "action", action: "review", feature: "auth" }),
@@ -55,6 +58,60 @@ test("actionToCommand maps hub actions to skill commands", () => {
 	);
 });
 
+test("implementationCommand creates fresh-session command invocations", () => {
+	assert.equal(implementationCommand("auth"), "/iterator-implement auth");
+	assert.equal(
+		implementationCommand("auth", { auto: true, guidance: "keep tests red" }),
+		"/iterator-implement auth --auto — keep tests red",
+	);
+});
+
+test("implementationHandoffState preserves auto and wave lifecycle state only for new sessions", () => {
+	const featureWave = { queue: ["b"], active: "a", results: [] };
+	const entries = [
+		{ type: "custom", customType: "other", data: {} },
+		{
+			type: "custom",
+			customType: "iterator-implementation-handoff",
+			data: { feature: "a", auto: true, autoSteps: 17, featureWave },
+		},
+	];
+	assert.deepEqual(implementationHandoffState(entries, "new"), {
+		feature: "a",
+		auto: true,
+		autoSteps: 17,
+		featureWave,
+	});
+	assert.equal(implementationHandoffState(entries, "reload"), null);
+	assert.equal(implementationHandoffState([], "new"), null);
+});
+
+test("implementationHandoffState defaults malformed circuit-breaker state", () => {
+	const entries = [
+		{
+			type: "custom",
+			customType: "iterator-implementation-handoff",
+			data: { feature: "a", autoSteps: -1 },
+		},
+	];
+	assert.equal(implementationHandoffState(entries, "new").autoSteps, 0);
+});
+
+test("shouldApplyRole lets fresh auto and wave implementers own their role", () => {
+	assert.equal(shouldApplyRole("implementer", { mode: "auto" }), true);
+	assert.equal(
+		shouldApplyRole("implementer", {
+			mode: "manual",
+			featureWave: { active: "auth" },
+		}),
+		true,
+	);
+	assert.equal(
+		shouldApplyRole("tester", { mode: "manual", featureWave: null }),
+		true,
+	);
+});
+
 test("actionToCommand returns null for cancel/timeout/garbage", () => {
 	assert.equal(actionToCommand({ type: "cancel" }), null);
 	assert.equal(actionToCommand({ type: "timeout" }), null);
@@ -186,7 +243,11 @@ test("composeAmbientContext builds the state line and anchored-knowledge list",
 		plan: { title: "Add JWT auth", status: "approved" },
 		progress: { done: 3, total: 7 },
 		features: [
-			{ name: "auth-middleware", testsStatus: "red" },
+			{
+				name: "auth-middleware",
+				testsStatus: "red",
+				tests: ["test/auth.test.mjs", "test/auth-policy.test.mjs"],
+			},
 			{ name: "config-module", testsStatus: "green" },
 		],
 	};
@@ -202,7 +263,10 @@ test("composeAmbientContext builds the state line and anchored-knowledge list",
 	const out = composeAmbientContext(hub, implement, concepts);
 	assert.match(out, /Plan "Add JWT auth" — 3\/7 features done/);
 	assert.match(out, /next ready: auth-middleware/);
-	assert.match(out, /tests red: auth-middleware/);
+	assert.match(
+		out,
+		/committed red tests: auth-middleware \(test\/auth\.test\.mjs, test\/auth-policy\.test\.mjs\)/,
+	);
 	assert.match(
 		out,
 		/\[pitfalls\/token-clock-skew\] JWT clock skew — Fresh tokens fail without leeway\. \(memory\/pitfalls\/token-clock-skew\.md\)/,
diff --git a/test/session-server.test.mjs b/test/session-server.test.mjs
index 4d00ebe..ac012a0 100644
--- a/test/session-server.test.mjs
+++ b/test/session-server.test.mjs
@@ -44,6 +44,33 @@ const viewHtml = (marker) =>
 	`<!DOCTYPE html><html><body>${marker} run=${srvMod.RUN_ID}</body></html>`;
 
 /** Read the first SSE event from /events. */
+function sseEvent(origin, wanted) {
+	return new Promise((resolve, reject) => {
+		const req = http.get(`${origin}/events`, (res) => {
+			let buf = "";
+			res.on("data", (d) => {
+				buf += d;
+				for (const packet of buf.split("\n\n")) {
+					const m = packet.match(/event: (\w+)\ndata: (.*)/);
+					if (!m || m[1] !== wanted) continue;
+					try {
+						const data = JSON.parse(m[2]);
+						req.destroy();
+						resolve({ event: m[1], data });
+						return;
+					} catch (error) {
+						req.destroy();
+						reject(error);
+						return;
+					}
+				}
+			});
+		});
+		req.on("error", () => {});
+		setTimeout(() => reject(new Error(`no ${wanted} SSE event`)), 3000).unref();
+	});
+}
+
 function firstSseEvent(origin) {
 	return new Promise((resolve, reject) => {
 		const req = http.get(`${origin}/events`, (res) => {
@@ -282,11 +309,40 @@ test("backlog writes remain available while an agent is working", async () => {
 			action: "create",
 			title: "Next idea",
 		});
+		const bulk = await fetch(`${origin}/submit?r=${srvMod.RUN_ID}`, {
+			method: "POST",
+			body: JSON.stringify({
+				type: "backlog",
+				action: "select-many",
+				ids: ["first-idea", "second-bug"],
+				selected: true,
+			}),
+		});
+		assert.equal(bulk.status, 200);
+		await sleep(20);
+		assert.deepEqual(unsolicited, {
+			type: "backlog",
+			action: "select-many",
+			ids: ["first-idea", "second-bug"],
+			selected: true,
+		});
 		assert.equal(
 			session.isWorking(),
 			true,
 			"backlog writes preserve the active model guard",
 		);
+		// The extension refreshes both tabs after the deterministic write. A
+		// refresh while working must update stored Planning HTML without releasing
+		// or replacing the active claim.
+		session.showView({
+			step: "planning",
+			render: () => viewHtml("PLANNING-WITH-SAVED-CANDIDATE"),
+		});
+		assert.match(
+			await (await fetch(`${origin}/view?tab=planning`)).text(),
+			/PLANNING-WITH-SAVED-CANDIDATE/,
+		);
+		assert.equal(session.isWorking(), true);
 		const blocked = await fetch(`${origin}/submit?r=${srvMod.RUN_ID}`, {
 			method: "POST",
 			body: '{"type":"action","action":"implement","feature":"other"}',
@@ -551,6 +607,92 @@ test("stop() resolves a pending round, frees the port, and removes the registry
 	);
 });
 
+test("shell-owned Settings preserves the active tab and replays on reconnect", async () => {
+	const { session, origin } = await startSession();
+	try {
+		session.showView({
+			step: "planning",
+			render: () => viewHtml("PLANNING-UNDER-MODAL"),
+			activate: true,
+		});
+		session.showModal({ render: () => viewHtml("SETTINGS-MODAL") });
+		assert.match(
+			await (await fetch(origin + "/settings")).text(),
+			/SETTINGS-MODAL/,
+		);
+		assert.match(
+			await (await fetch(origin + "/view?tab=planning")).text(),
+			/PLANNING-UNDER-MODAL/,
+		);
+		assert.match(
+			await (await fetch(origin + "/")).text(),
+			/let tab = "planning"/,
+		);
+		const replay = await sseEvent(origin, "modal");
+		assert.deepEqual(replay.data, { open: true, v: 1 });
+		assert.equal(session.closeModal(), true);
+		assert.match(
+			await (await fetch(origin + "/settings")).text(),
+			/waiting for the next step/,
+		);
+	} finally {
+		await session.stop();
+	}
+});
+
+test("Settings submit does not settle the pending round beneath the modal", async () => {
+	let unsolicited = null;
+	const { session, origin } = await startSession({
+		onUnsolicited: (result) => (unsolicited = result),
+	});
+	try {
+		const round = session.showStep({
+			step: "review",
+			render: () => viewHtml("REVIEW"),
+		});
+		session.showModal({ render: () => viewHtml("SETTINGS") });
+		const response = await fetch(`${origin}/submit`, {
+			method: "POST",
+			body: '{"type":"settings","values":{"auto_mode":"on"}}',
+		});
+		assert.equal(response.status, 200);
+		await sleep(20);
+		assert.deepEqual(unsolicited, {
+			type: "settings",
+			values: { auto_mode: "on" },
+		});
+		assert.equal(session.hasPending(), true);
+		await fetch(`${origin}/submit?r=${srvMod.RUN_ID}`, {
+			method: "POST",
+			body: '{"type":"review-approved"}',
+		});
+		assert.deepEqual(await round, { type: "review-approved" });
+	} finally {
+		await session.stop();
+	}
+});
+
+test("Settings dismissal remains available while Work is blocked", async () => {
+	let unsolicited = null;
+	const { session, origin } = await startSession({
+		onUnsolicited: (result) => (unsolicited = result),
+	});
+	try {
+		session.showView({ step: "hub", render: () => viewHtml("WORK") });
+		session.showWorking("AI is working…");
+		const response = await fetch(`${origin}/submit`, {
+			method: "POST",
+			body: '{"type":"settings-close"}',
+		});
+		assert.equal(response.status, 200);
+		await sleep(20);
+		assert.deepEqual(unsolicited, { type: "settings-close" });
+		assert.equal(session.isWorking(), true);
+	} finally {
+		await session.stop();
+	}
+});
+
 test("showView can intentionally activate Planning for the startup landing page", async () => {
 	const { session, origin } = await startSession();
 	try {
@@ -577,6 +719,23 @@ test("showView can intentionally activate Planning for the startup landing page"
 	}
 });
 
+test("showView can intentionally activate Work after an approved transition", async () => {
+	const { session, origin } = await startSession();
+	try {
+		session.showView({
+			step: "hub",
+			render: () => viewHtml("ACTIVE-WORK"),
+			activate: true,
+		});
+		const sse = await firstSseEvent(origin);
+		assert.equal(sse.event, "view");
+		assert.equal(sse.data.tab, "work");
+		assert.match(await (await fetch(origin + "/")).text(), /let tab = "work"/);
+	} finally {
+		await session.stop();
+	}
+});
+
 test("tabs: steps render into their tab; inactive-tab refreshes are stored silently", async () => {
 	const { session, origin } = await startSession();
 	try {
@@ -703,6 +862,39 @@ test("an idle submit from a stale-run view still dispatches as unsolicited", asy
 	}
 });
 
+test("an idle structured plan is forwarded intact for direct review", async () => {
+	let unsolicited = null;
+	const { session, origin } = await startSession({
+		onUnsolicited: (r) => (unsolicited = r),
+	});
+	try {
+		session.showView({ step: "planning", render: () => viewHtml("PLANNING") });
+		const payload = {
+			type: "action",
+			action: "plan-fast-track",
+			prompt: "# Complete plan",
+			structuredPlan: {
+				title: "Complete plan",
+				plan: {
+					goal: "Ship it.",
+					architecture: "- Existing seams.",
+					keyDecisions: "- Keep review.",
+				},
+				dependencies: [],
+			},
+		};
+		const res = await fetch(`${origin}/submit?r=${srvMod.RUN_ID}`, {
+			method: "POST",
+			body: JSON.stringify(payload),
+		});
+		assert.equal(res.status, 200);
+		await sleep(20);
+		assert.deepEqual(unsolicited, payload);
+	} finally {
+		await session.stop();
+	}
+});
+
 test("an idle backlog submission is forwarded without unrelated dashboard state", async () => {
 	let unsolicited = null;
 	const { session, origin } = await startSession({
diff --git a/test/ui.test.mjs b/test/ui.test.mjs
index 65eeb48..abcd8ec 100644
--- a/test/ui.test.mjs
+++ b/test/ui.test.mjs
@@ -128,10 +128,35 @@ test("shared client JS wires read-only mode while the agent works", () => {
 	);
 });
 
-test("mdToHtml refuses javascript: links", () => {
-	const html = renderPage({ step: "t", data: {}, body: "", clientJs: "" });
-	// The linkify branch must be guarded by a protocol whitelist.
-	assert.ok(html.includes("https?:|mailto:"));
+test("mdToHtml preserves query strings while escaping href quotes", () => {
+	const html = renderPage({
+		step: "t",
+		data: {},
+		body: "",
+		clientJs: `globalThis.renderedLink = mdToHtml('[safe](https://example.test/?a=1&b="two")');`,
+	});
+	const script = html.match(/<script>\n([\s\S]*?)\n<\/script>/)[1];
+	const branch = { textContent: "" };
+	const classList = { contains: () => false, toggle: () => {} };
+	const window = { addEventListener: () => {}, close: () => {} };
+	window.parent = window;
+	const ctx = vm.createContext({
+		window,
+		document: {
+			documentElement: { dataset: { theme: "dark" } },
+			body: { classList },
+			getElementById: (id) => (id === "branch" ? branch : null),
+		},
+		navigator: { sendBeacon: () => true },
+		fetch: async () => ({ status: 200 }),
+		alert: () => {},
+	});
+	vm.runInContext(script, ctx);
+	assert.match(
+		ctx.renderedLink,
+		/href="https:\/\/example\.test\/\?a=1&amp;b=&quot;two&quot;"/,
+	);
+	assert.doesNotMatch(ctx.renderedLink, /&amp;amp;/);
 });
 
 test("CSS exports are non-empty and themed", () => {
@@ -313,8 +338,11 @@ test("plan-less Planning exposes init and plan actions without losing the goal",
 	assert.match(html, /Initialize memory/);
 	assert.match(html, /Create plan/);
 	assert.match(html, /action:'iterator-init'/);
-	assert.match(html, /action:'plan'/);
-	assert.match(html, /prompt: goal\.value\.trim\(\) \|\| null/);
+	assert.match(html, /action:structuredPlan \? 'plan-fast-track' : 'plan'/);
+	assert.match(html, /const prompt = goal\.value\.trim\(\) \|\| null/);
+	assert.match(html, /function parseStructuredPlan/);
+	assert.match(html, /found < 3/);
+	assert.match(html, /plan:\{goal, architecture, keyDecisions\}/);
 	assert.match(html, /PLAN_DRAFT_KEY/);
 	assert.match(html, /if\(D\.plan\) clearPlanDraft\(\)/);
 	const initHandler = html.slice(
@@ -372,6 +400,18 @@ test("planning backlog submits scoped CRUD actions and hands selected candidates
 	);
 	assert.match(html, /Plan selected candidates/);
 	assert.match(html, /selectedBacklogGoal/);
+	assert.match(html, /<div class="at-wrap"><textarea/);
+	assert.match(html, /wireAtMenu\(details, detailsMenu\)/);
+	assert.match(html, /data-kind="all">All/);
+	assert.match(html, /data-kind="idea">Ideas/);
+	assert.match(html, /data-kind="bug">Bugs/);
+	assert.match(html, /Select visible/);
+	assert.match(html, /Deselect visible/);
+	assert.match(html, /const visibleItems =/);
+	assert.match(html, /action:'select-many'/);
+	assert.match(html, /ids:targets\.map\(item => item\.id\)/);
+	assert.doesNotMatch(html, /for\(const item of targets\)\{\s*const saved/);
+	assert.match(html, /refreshHandoff/);
 	// Long history feeds scroll within their own region; headings and controls
 	// remain available outside it.
 	assert.match(html, /backlog-list bounded-list/);
@@ -446,15 +486,95 @@ test("hub gates Implement/Review on status and renders escalation + review-plan
 	assert.match(html, /Review all/);
 	assert.match(html, /action\('review-all'/);
 	assert.match(html, /Array\.isArray\(D\.reviewWave\)/);
-	// Active feature context and management live on Work.
+	// Active feature and plan lifecycle management live on Work.
 	assert.match(html, /renderGraphInto/);
 	assert.match(html, /action\('cancel-feature'/);
-	// Plan-lifecycle controls live on the Planning surface, not Work.
-	assert.doesNotMatch(html, /action\('review-plan'/);
-	assert.doesNotMatch(html, /Retires the plan/);
+	assert.match(html, /action\('review-plan'/);
+	assert.match(html, /Retires the plan/);
+	assert.match(html, /action\('cancel-plan'/);
+});
+
+test("Work routes decision conflicts through reviewed knowledge updates", async () => {
+	const { render: hub } = await import("../lib/views/hub.mjs");
+	const html = hub({
+		step: "hub",
+		branch: "iterator/p",
+		plan: { title: "P", status: "approved" },
+		stage: "implementing",
+		progress: { done: 0, total: 1 },
+		features: [
+			{
+				name: "orm-feature",
+				title: "ORM feature",
+				description: "Uses an ORM",
+				status: "pending",
+				size: "small",
+				testsStatus: "none",
+				dependsOn: [],
+				ready: true,
+				conflicts: [
+					{
+						decision: "decisions/no-orm",
+						title: "No ORM",
+						note: "Introduces an ORM",
+						files: ["src/db/*.ts"],
+					},
+				],
+			},
+		],
+		readyWave: [],
+		reviewWave: [],
+		state: {},
+		dirty: { count: 0, files: [] },
+		retired: [],
+	});
+	assert.match(html, /Review decision/);
+	assert.match(html, /resolve-memory-conflict/);
+	assert.match(html, /target:conflict\.decision/);
+	assert.match(html, /anchors:Array\.isArray\(conflict\.files\)/);
+	assert.match(html, /review it before implementation/);
 });
 
-test("planning drives plan-lifecycle controls from the server-derived stage", async () => {
+test("Work presents committed red tests as the implementation target", async () => {
+	const { render: hub } = await import("../lib/views/hub.mjs");
+	const html = hub({
+		step: "hub",
+		branch: "iterator/p",
+		plan: { title: "P", status: "approved" },
+		stage: "implementing",
+		progress: { done: 0, total: 1 },
+		features: [
+			{
+				name: "auth",
+				title: "Auth",
+				status: "pending",
+				size: "small",
+				testsStatus: "red",
+				tests: ["test/auth.test.mjs"],
+				testCount: 1,
+				dependsOn: [],
+				ready: true,
+				waitingOn: [],
+				hasDiff: false,
+				hasCommits: true,
+				conflicts: 0,
+			},
+		],
+		state: { mode: "manual", paused: false, phase: "idle", strikes: {} },
+		settings: {},
+		dirty: { count: 0, files: [] },
+		retired: [],
+		backlog: [],
+	});
+	assert.match(html, /tests committed · intentionally red/);
+	assert.match(html, /Committed red tests — implementation target/);
+	assert.match(html, /c\.tests\.map/);
+	assert.match(html, /Drive tests green/);
+	assert.match(html, /Tests committed \(red\)/);
+	assert.match(html, /Committed red tests are the implementation target/);
+});
+
+test("planning keeps active work staged while Work drives lifecycle controls", async () => {
 	const { render: planning } = await import("../lib/views/planning.mjs");
 	const html = planning({
 		step: "planning",
@@ -496,15 +616,15 @@ test("planning drives plan-lifecycle controls from the server-derived stage", as
 		],
 		backlog: [],
 	});
-	// Lifecycle buttons key off the server-derived stage.
-	assert.match(html, /D\.stage==='retirable'/);
-	assert.match(html, /action\('review-plan'/);
-	assert.match(html, /Retires the plan/);
-	assert.match(html, /action\('cancel-plan'/);
-	// Active features and their dependency graph live on Work, not Planning.
+	assert.match(html, /Planning is staged/);
+	assert.match(html, /Open Work/);
+	assert.match(html, /action\('hub'/);
+	// Active lifecycle, features, and execution controls are absent from Planning.
+	assert.doesNotMatch(html, /action\('review-plan'/);
+	assert.doesNotMatch(html, /Retires the plan/);
+	assert.doesNotMatch(html, /action\('cancel-plan'/);
 	assert.doesNotMatch(html, /action\('cancel-feature'/);
 	assert.doesNotMatch(html, /renderGraphInto/);
-	// The execution controls live on Work, not here.
 	assert.doesNotMatch(html, /action\('implement'/);
 	assert.doesNotMatch(html, /auto-implement/);
 	// Retired-plan browsing remains a planning concern.
@@ -580,14 +700,82 @@ test("planning hero goal box persists an unsent draft and clears it on plan star
 	assert.match(html, /iterator:plan-goal-draft/);
 	assert.match(html, /localStorage\.getItem\(PLAN_DRAFT_KEY\)/);
 	assert.match(html, /goal\.addEventListener\('input', saveDraft\)/);
-	// Direct planning clears on accepted submit; initialization retains the goal
-	// until the continued plan actually appears.
-	assert.match(html, /if\(__submitted\) clearPlanDraft\(\)/);
+	// Goal-only planning clears after dispatch as before. A complete plan keeps
+	// its draft through browser review and clears when the approved plan appears.
+	assert.match(html, /if\(__submitted && !structuredPlan\) clearPlanDraft\(\)/);
 	assert.match(html, /if\(D\.plan\) clearPlanDraft\(\)/);
 	// The larger input within the saved design parameters.
 	assert.match(html, /textarea\.goal\{[^}]*min-height:132px/);
 });
 
+test("Settings closes through the shell modal without a cancellation beacon", async () => {
+	const { render: settings } = await import("../lib/views/settings.mjs");
+	const html = settings({
+		branch: "main",
+		plan: "P",
+		defined: true,
+		settings: {},
+	});
+	assert.doesNotMatch(html, /onclick="cancelFlow\(\)"/);
+	assert.match(html, /type:'settings-close'/);
+	assert.match(html, /allowWhileWorking:true/);
+	assert.match(html, /id="primary" onclick="primaryClick\(\)">Close/);
+});
+
+test("memory review highlights changed markdown sections without changing verdicts", async () => {
+	const { render } = await import("../lib/views/memory-review.mjs");
+	const html = render({
+		branch: "main",
+		mode: "consolidate",
+		round: 1,
+		areas: [{ name: "patterns", description: "Patterns" }],
+		memories: [
+			{
+				id: "patterns/safe",
+				area: "patterns",
+				action: "update",
+				title: "Safe rendering",
+				type: "Pattern",
+				reason: "Resolve the Work decision conflict.",
+				body: "# Kept\nSame.\n\n# Changed\nNew <value>.",
+				existingBody:
+					"# Kept\nSame.\n\n# Changed\nOld value.\n\n# Removed\nGone.",
+			},
+		],
+	});
+	assert.match(html, /function markdownBlocks/);
+	assert.match(html, /function markdownChangeHtml/);
+	assert.match(html, /Changed section/);
+	assert.match(html, /Removed section/);
+	assert.match(html, /Why this change:/);
+	assert.match(html, /Resolve the Work decision conflict\./);
+	assert.match(html, /change-block\.added,\.change-block\.modified/);
+	assert.match(html, /m\.action === 'create' \|\| m\.action === 'update'/);
+	assert.match(html, /type: 'review-approved'/);
+	assert.match(html, /decisions: decisions/);
+	assert.match(html, /\\u003cvalue>/, "embedded proposal data stays inert");
+});
+
+test("structured plan review accepts direct edits without a planner feedback round", async () => {
+	const { render: plan } = await import("../lib/views/plan.mjs");
+	const html = plan({
+		branch: "main",
+		title: "Complete plan",
+		fastTrack: true,
+		plan: {
+			goal: "Ship it.",
+			architecture: "- Use existing seams.",
+			keyDecisions: "- Keep approval.",
+		},
+		dependencies: [],
+	});
+	assert.match(html, /Structured plan · ready for review/);
+	assert.match(html, /if\(D\.fastTrack\)/);
+	assert.match(html, /if\(D\.fastTrack\) return false/);
+	assert.match(html, /type: changed \? 'plan-feedback' : 'plan-approved'/);
+	assert.match(html, /body\.fast-plan \.cmt-btn/);
+});
+
 test("idle dashboard tabs omit the header Cancel button; round views keep it with a tooltip", async () => {
 	const CANCEL_BTN = 'onclick="cancelFlow()"';
 	const { render: hub } = await import("../lib/views/hub.mjs");
diff --git a/test/write.test.mjs b/test/write.test.mjs
index 07de4cc..032fc48 100644
--- a/test/write.test.mjs
+++ b/test/write.test.mjs
@@ -3215,6 +3215,73 @@ test("loadBundle re-roots every gather/write into the plan's worktree", async ()
 // ---------------------------------------------------------------------------
 // op: backlog
 
+test("backlog writer atomically persists visible-set selections", () => {
+	const root = makeRepo();
+	try {
+		const idea = applyOp(
+			{
+				op: "backlog",
+				action: "create",
+				kind: "idea",
+				title: "Idea",
+				details: "",
+			},
+			root,
+		).item;
+		const bug = applyOp(
+			{
+				op: "backlog",
+				action: "create",
+				kind: "bug",
+				title: "Bug",
+				details: "",
+			},
+			root,
+		).item;
+		let result = applyOp(
+			{
+				op: "backlog",
+				action: "select-many",
+				ids: [idea.id, bug.id],
+				selected: true,
+			},
+			root,
+		);
+		assert.deepEqual(
+			result.changedItems.map((item) => item.id),
+			[idea.id, bug.id],
+		);
+		assert.ok(result.items.every((item) => item.selected));
+		result = applyOp(
+			{ op: "backlog", action: "select", id: idea.id, selected: false },
+			root,
+		);
+		assert.equal(
+			result.items.find((item) => item.id === idea.id).selected,
+			false,
+		);
+		assert.equal(
+			result.items.find((item) => item.id === bug.id).selected,
+			true,
+		);
+		assert.throws(
+			() =>
+				applyOp(
+					{
+						op: "backlog",
+						action: "select-many",
+						ids: [idea.id, "missing"],
+						selected: false,
+					},
+					root,
+				),
+			/no backlog item 'missing'/,
+		);
+	} finally {
+		rmSync(root, { recursive: true, force: true });
+	}
+});
+
 test("backlog writer saves, validates, selects, edits, and deletes candidates", () => {
 	const root = makeRepo();
 	try {