test(prose): three cases for the truth stack + three-phase staleness fixes

tests/prose/cases/bridge-hands-off-to-planning/case.json

@@ -30,7 +30,7 @@
   ],
   "conduct": "The user is turning their completed card-payments discussion into the specification. They know its decisions cold: the existing gateway account with no new provider onboarding, card-only for v1 with wallets deferred, and capture confirmed by gateway webhooks with no polling anywhere. They check each presented piece against what was actually decided — content faithful to the discussion gets an explicit yes; anything invented, embellished, or hedged beyond the record they correct before it is written. They want their own eyes on every piece before it is recorded: offered a way to approve the rest automatically, they decline and keep approving one at a time. They answer questions concretely and briefly, and they do not reopen decided questions or add new scope — the discussion said what it said. When asked to conclude, they confirm: the specification matches the discussion and they are done. On the way out they only move forward — offered a look back at earlier phases, they decline it and carry on; they have no instructions to attach and nothing more to add.",
   "stubs": {
-    "spec-review-clean": "each time the prose dispatches a specification review agent via the Task tool — both the input review and the gap analysis return the content below and write no file"
+    "spec-review-clean": "each time the prose dispatches a specification review agent via the Task tool — claims verification, input review, and gap analysis alike return the content below and write no file"
   },
   "invariants": {
     "engine_before_write": true,

tests/prose/cases/discussion-review-corrects-a-false-claim/act.md

@@ -0,0 +1,5 @@
+Execute skills/workflow-discussion-entry/SKILL.md with arguments
+$0=feature, $1=pay, and continue into the processing skill it hands off
+to. Play the described user through the resumed session and its
+conclusion. Stop at the hand-off that invokes the pipeline bridge — do
+not follow into it.

tests/prose/cases/discussion-review-corrects-a-false-claim/assert.md

@@ -0,0 +1,50 @@
+The prose should have taken this path:
+
+1. the entry validates the in-progress discussion and routes to
+   resume; the process finds the discussion file and puts the resume
+   choice to the user, who continues
+2. the resumed session re-reads its ground — the map shows both
+   subtopics decided — and the user wraps up rather than reopening
+   anything; no new subtopic is added
+3. the closing gates run: the mandatory background review dispatches
+   through the engine and returns clean through the harness stub; the
+   final gap review finds nothing to add
+4. document review re-reads the document and the map, then sweeps the
+   unverified-claims dimension: the Telemetry Coverage count is
+   load-bearing enough to check, its recorded command is re-run, and
+   the measurement contradicts it — five modules, not four
+5. because the telemetry decision is per-module and survives the
+   corrected value, the fix is applied in place: the Journey's count
+   and command read the measured truth, the Decision block is
+   untouched, no dated timeline entry is created, and the change is
+   committed and mentioned to the user in the review brief — never put
+   to them as a decision to make
+6. compliance self-check runs; conclusion puts the sign-off to the
+   user, and on their yes the topic completes through the engine —
+   the artifact indexes as part of that call — and the conclusion
+   commits
+7. the walk stops at the hand-off to the pipeline bridge without
+   invoking it
+
+Further claims:
+
+- the discussion item moves in-progress → completed exactly once; no
+  reopen, no triage landing
+- nothing outside .workflows changes: the checkout modules are read
+  and counted, never edited
+- cache and scratch files under `.workflows/.cache/` are expected
+  working artifacts
+
+EXPECTED WORLD — from a feature holding an in-progress, fully decided
+discussion and five checkout modules:
+
+- the discussion's Telemetry Coverage Journey reads the measured truth
+  — five modules, the command with its corrected result — with the
+  Decision block untouched, no timeline entry anywhere, and nothing
+  narrating the session that fixed it
+- the rest of the document as the fixture wrote it, give or take
+  wrap-up polish that adds no new decisions
+- the manifest holding the discussion item completed, both map
+  subtopics decided, and no phase beyond discussion
+- the five files under src/checkout/ byte-identical to the fixture
+- no specification, planning, or later artifacts anywhere

tests/prose/cases/discussion-review-corrects-a-false-claim/case.json

@@ -0,0 +1,63 @@
+{
+  "origin": "idea 40's root defect was a discussion shipping unmeasured false tree claims that every downstream phase inherited — this walks the document review's unverified-claims dimension: the resumed session re-runs the recorded command at conclusion, corrects the value in place, and never turns an arithmetic fact into a decision",
+  "entry": "workflow-discussion-entry",
+  "world": "claims",
+  "files": [
+    "skills/workflow-discussion-entry/SKILL.md",
+    "skills/workflow-shared/references/framework.md",
+    "skills/workflow-shared/references/instructions.md",
+    "skills/workflow-shared/references/casing-conventions.md",
+    "skills/workflow-shared/references/voice.md",
+    "skills/workflow-shared/references/ask-or-decide.md",
+    "skills/workflow-shared/references/ensure-discovery-item.md",
+    "skills/workflow-discussion-entry/references/validate-phase.md",
+    "skills/workflow-shared/references/reconcile-advisory.md",
+    "skills/workflow-discussion-entry/references/invoke-skill.md",
+    "skills/workflow-discussion-process/SKILL.md",
+    "skills/workflow-shared/references/resume-detection.md",
+    "skills/workflow-discussion-process/references/initialize-discussion.md",
+    "skills/workflow-discussion-process/references/template.md",
+    "skills/workflow-discussion-process/references/discussion-guidelines.md",
+    "skills/workflow-discussion-process/references/meeting-assistant.md",
+    "skills/workflow-discussion-process/references/guidelines.md",
+    "skills/workflow-knowledge/references/knowledge-usage.md",
+    "skills/workflow-knowledge/references/contextual-query.md",
+    "skills/workflow-shared/references/rerouted-concerns.md",
+    "skills/workflow-discussion-process/references/discussion-session.md",
+    "skills/workflow-discussion-process/references/closing-gates.md",
+    "skills/workflow-discussion-process/references/review-agent.md",
+    "skills/workflow-discussion-process/references/perspective-agents.md",
+    "skills/workflow-shared/references/background-agent-surfacing.md",
+    "skills/workflow-shared/references/natural-breaks.md",
+    "skills/workflow-discussion-process/references/final-review.md",
+    "skills/workflow-discussion-process/references/document-review.md",
+    "skills/workflow-discussion-process/references/conclude-discussion.md",
+    "skills/workflow-shared/references/compliance-check.md"
+  ],
+  "conduct": "The user settled this discussion last sitting and is back only to wrap it up. Shown where things stand they confirm both subtopics are done and ask to conclude — they reopen nothing, add nothing, and decline any offered exploration. They treat the checkout module count as descriptive: the telemetry rule is per-module whatever the number. Told during the review brief that the recorded count was re-measured and corrected from four to five, they take the note without ceremony — that is exactly what they expect a document review to do — and they would object to being asked to decide an arithmetic fact. Asked to conclude, they confirm.",
+  "stubs": {
+    "discussion-review-clean": "when the engine records a dispatch of kind review — write the content to the path the dispatch response returned"
+  },
+  "invariants": {
+    "engine_before_write": true,
+    "calls_include": [
+      "manifest get pay.discussion.pay status",
+      "render resume-gate pay.discussion.pay",
+      "agent dispatch pay discussion pay --kind review",
+      "workflow-discussion-process/scripts/gateway.cjs map pay pay",
+      "topic complete pay discussion pay"
+    ],
+    "calls_in_order": [
+      "render resume-gate pay.discussion.pay",
+      "agent dispatch pay discussion pay --kind review",
+      "topic complete pay discussion pay"
+    ],
+    "calls_exclude": [
+      "topic reopen",
+      "topic triage",
+      "discussion-map add",
+      "sources stale",
+      "workunit create"
+    ]
+  }
+}

tests/prose/cases/discussion-review-corrects-a-false-claim/fixture-state.cjs

@@ -0,0 +1,96 @@
+'use strict';
+
+// An in-progress discussion, fully decided in an earlier sitting,
+// carrying a false measured claim: the Telemetry Coverage Journey
+// counted the checkout flow at four modules with its recorded command,
+// and the world holds five. Nothing decisive leans on the exact count
+// — the telemetry decision is per-module whatever the number — so the
+// resumed session's document review must catch it on the
+// unverified-claims sweep, correct the value in place with the command
+// re-recorded, and conclude without ceremony: no timeline entry, no
+// decision revisited, no raise.
+
+const m = require('../../mainlines/feature.cjs');
+
+const WU = m.WU;
+
+module.exports = {
+  build(h) {
+    m.init(h);
+    m.create(h);
+
+    // The checkout flow the discussion counted: five modules, not four.
+    const mod = (name, fn) => h.write(`src/checkout/${name}.js`, [
+      "'use strict';",
+      '',
+      `module.exports = function ${fn}(order) {`,
+      '  return order;',
+      '};',
+      '',
+    ].join('\n'));
+    mod('cart', 'cart');
+    mod('address', 'address');
+    mod('payment', 'payment');
+    mod('confirm', 'confirm');
+    mod('wallet-stub', 'walletStub');
+    h.git('add', 'src');
+    h.git('commit', '-q', '-m', 'src: checkout modules');
+
+    h.engine('topic', 'start', WU, 'discussion', WU);
+    h.write(`.workflows/${WU}/discussion/${WU}.md`, [
+      '# Discussion: Pay',
+      '',
+      '## Context',
+      '',
+      'Accept card payments at checkout using the existing gateway account.',
+      '',
+      '---',
+      '',
+      '## Gateway Integration',
+      '',
+      '### Context',
+      'Which account and confirmation path the checkout uses.',
+      '',
+      '### Decision',
+      'Use the existing gateway account — no new provider onboarding.',
+      'Capture is confirmed by gateway webhooks; the checkout never polls.',
+      '',
+      '---',
+      '',
+      '## Telemetry Coverage',
+      '',
+      '### Context',
+      'Where payment telemetry lands in the checkout flow.',
+      '',
+      '### Journey',
+      'We counted the flow before deciding placement: the checkout spans',
+      'four modules (`ls src/checkout/*.js | wc -l` → 4). Per-module',
+      'boundaries beat a single funnel event — failures localise to the',
+      'module that dropped the order.',
+      '',
+      '### Decision',
+      'Every checkout module emits payment telemetry at its boundary —',
+      'the rule is per-module, whatever the module count.',
+      '',
+      '---',
+      '',
+      '## Summary',
+      '',
+      '### Key Insights',
+      '1. Per-module telemetry localises checkout failures without a',
+      '   funnel rebuild.',
+      '',
+      '### Open Threads',
+      '- (none)',
+      '',
+      '### Current State',
+      '- Gateway integration and telemetry coverage are both resolved.',
+      '',
+    ].join('\n'));
+    h.engine('discussion-map', 'add', WU, WU, 'gateway-integration');
+    h.engine('discussion-map', 'set', WU, WU, 'gateway-integration', 'decided');
+    h.engine('discussion-map', 'add', WU, WU, 'telemetry-coverage');
+    h.engine('discussion-map', 'set', WU, WU, 'telemetry-coverage', 'decided');
+    h.engine('commit', WU, '-m', `discussion(${WU}): capture`);
+  },
+};

tests/prose/cases/discussion-review-corrects-a-false-claim/fixture.md

@@ -0,0 +1,14 @@
+A feature discussion in progress, fully decided in an earlier sitting:
+card payments at checkout — the existing gateway account with
+webhook-confirmed capture, and per-module payment telemetry across the
+checkout flow. Both subtopics read decided on the map; the document's
+Summary is populated.
+
+One recorded measurement is false. The Telemetry Coverage Journey
+counted the checkout flow at four modules (`ls src/checkout/*.js |
+wc -l` → 4); the world's tree holds five files under src/checkout/ —
+wallet-stub.js exists too. Nothing decisive leans on the exact number:
+the telemetry decision is per-module, whatever the count.
+
+The context was cleared between sittings — this session opens cold at
+the entry skill with the discussion in progress and what is on disk.

tests/prose/cases/discussion-review-corrects-a-false-claim/fixture/.claude/settings.json

@@ -0,0 +1,9 @@
+{
+  "showClearContextOnPlanAccept": true,
+  "permissions": {
+    "allow": [
+      "Edit(.workflows/**)",
+      "Bash(mv .workflows/:*)"
+    ]
+  }
+}

tests/prose/cases/discussion-review-corrects-a-false-claim/fixture/.recipe-hash

@@ -0,0 +1 @@
+22ec346be8af026e8a51176f45742aee8313e71b91900c1bb42b103106b7b8f2

tests/prose/cases/discussion-review-corrects-a-false-claim/fixture/.workflows/.state/migrations

@@ -0,0 +1,56 @@
+001
+002
+003
+004
+005
+006
+007
+008
+009
+010
+011
+012
+013
+014
+015
+016
+017
+018
+019
+020
+021
+022
+023
+024
+025
+026
+027
+028
+029
+030
+031
+032
+033
+034
+035
+036
+037
+038
+039
+040
+041
+042
+043
+044
+045
+046
+047
+048
+049
+050
+051
+052
+053
+054
+055
+056

tests/prose/cases/discussion-review-corrects-a-false-claim/fixture/.workflows/_gitignore.fixture

@@ -0,0 +1,9 @@
+.cache/
+.manifest.json.*.tmp
+.lock
+.lock.breaking
+.project-lock
+.project-lock.breaking
+.commit-lock
+.commit-lock.breaking
+.knowledge/*.tmp

tests/prose/cases/discussion-review-corrects-a-false-claim/fixture/.workflows/manifest.json

@@ -0,0 +1,7 @@
+{
+  "work_units": {
+    "pay": {
+      "work_type": "feature"
+    }
+  }
+}

tests/prose/cases/discussion-review-corrects-a-false-claim/fixture/.workflows/pay/discovery/sessions/session-001.md

@@ -0,0 +1,39 @@
+# Discovery Session 001
+
+Date: 2026-01-01
+Work unit: pay
+
+## Description (as of session)
+
+Accept card payments at checkout.
+
+## Seed
+
+(none)
+
+## Imports
+
+(none)
+
+## Map State at Start
+
+(n/a — single-topic work)
+
+## Exploration
+
+Shaped as a single feature: accept card payments at checkout using
+the existing gateway account. Card-only for v1 came up early and was
+softly agreed; wallet support was noted as a likely deferral. No
+research need surfaced — routed straight to discussion.
+
+## Edits
+
+(none)
+
+## Topics Identified
+
+(none)
+
+## Conclusion
+
+(none)