Dogfood Parity 1: Daily Driver Baseline

[shiny-code-bot]

Finish Line

Current main can be used as the daily-driver code command for real Every Code work without falling back to the old pre-pivot binary for ordinary sessions.

Current Status

State: Active. Dogfood Parity 1 now has deterministic exec rails, local build-gate reliability, git/patch mutation safety, and local install/rebuild preflight confidence on main.

Done:

• Make code-exec-harness the Dogfood Parity 1 exec gate #405 established tools/code-exec-harness as the deterministic no-live-token P0 code exec --json smoke gate.
• PR Make exec harness a dogfood gate #406 made just harness-smoke validate the checkout-built code binary via fake /v1/responses fixtures.
• Add exec-harness safety scenarios for Dogfood Parity 1 #407 / PR Add exec harness safety scenarios #408 added deterministic safety coverage for shell command round-trip behavior and read-only sandbox write denial.
• PR Unblock build-fast from native WebRTC metadata fetch #409 closed Port build-fast cache mode from stale build-cache branch #396 by removing native WebRTC git dependency metadata fetches from the default local build graph; ./build-fast.sh and just harness-smoke passed for that gate-clearing slice.
• Dogfood Parity 1: git and patch mutation harness safety #410 / PR Add git patch exec harness safety scenarios #411 added deterministic git/patch mutation safety coverage: workspace file-content assertions, git-status assertions, apply_patch round-trip/file-change/tool-result evidence, workspace-write edit persistence, and read-only denied-file absence.
• Dogfood Parity 1: install binary replacement confidence #412 / PR Add local code rebuild preflight check #413 added install/binary replacement confidence: non-mutating just local-code-rebuild-preflight, deterministic just local-code-rebuild-preflight-check, explicit symlink rollback messaging, and guarded code --version verification only when PATH resolves to the rebuilt binary.
• The harness can now assert command-event streams, grouped event-count alternatives, model-facing function_call_output payloads, post-run workspace contents, and post-run git status --porcelain across the current fork-transition JSONL shapes.

Current finish-line gap:

• This is a real first rail, not daily-driver parity yet. Daily-driver parity still needs overlay behavior for Auto Drive Restore Auto Drive parity fixtures on Codex fork baseline #398, Code Bridge/browser Define Code Bridge and browser control parity boundary #399, Auto Review proof metrics Map Auto Review proof metrics onto Codex review primitives #400, and enough real-session UX/telemetry behavior to run Every Code work without falling back to the old pre-pivot binary.
• Native WebRTC is intentionally outside the default build graph for now; restore it later only as an explicit opt-in overlay so fresh checkout metadata remains reliable.

Next action:

• Add the next Dogfood Parity 1 child slice for one high-value overlay behavior. Prefer Auto Review proof metrics (Map Auto Review proof metrics onto Codex review primitives #400) if the goal is to reduce broad-review noise and get actionable review evidence, or Code Bridge/browser (Define Code Bridge and browser control parity boundary #399) if the goal is interactive app debugging parity.
• Keep fake-response or local deterministic harness scenarios ahead of implementation overlays where feasible.

Acceptance Criteria

• A deterministic harness command exists and targets code-rs/target/dev-fast/code, not whatever code happens to be on PATH.
• The fake /v1/responses scenarios are documented as the required no-token smoke suite.
• Basic code exec --json smoke, shell command round-trip, read-only write-denial, git/patch mutation safety, workspace content checks, git-status checks, and tool-result evidence scenarios exist so broken binary/tool-result/sandbox behavior fails clearly.
• The local install/link path for dogfooding is tested and documented enough that replacing the old Every Code binary is deliberate and reversible.
• Stabilize Every Code parity on Codex CLI fork baseline #397 points future sessions here as the Dogfood Parity 1 milestone.
• Follow-up child issues remain scoped to overlays and fixture concepts rather than copying Every Code code wholesale.

Next Action

Create and land the next overlay child slice, likely Auto Review proof metrics (#400) or Code Bridge/browser (#399), with the same pattern used by #405, #407, #410, and #412: focused branch, scoped review agent, ./build-fast.sh, and just harness-smoke when relevant.

[cbusillo]

#399 boundary slice landed in #415.

Status update:

• Code Bridge/browser evidence is now classified in docs/code-bridge-browser-parity.md.
• The ledger and feature inventory point to that boundary map.
• Define Code Bridge and browser control parity boundary #399 remains open for the required fake bridge metadata/server/control fixtures before implementation ports.

Next parity slice candidate: continue #399 with those fixtures, or move to the next child domain once the fixture order is confirmed.

[cbusillo]

#399 Codex-aligned boundary fixture slice landed in #417.

Status update:

• App-server now has a regression fixture proving Code Bridge-shaped control method names do not enter the Codex Desktop remoteControl/* JSON-RPC path.
• Existing remoteControl/status/read compatibility remains disabled/upstream-shaped.
• The Define Code Bridge and browser control parity boundary #399 doc now records the alignment rule: prefer existing Codex app-server/protocol/command shapes where similar, and preserve Every Code bridge behavior through minimal additive overlays for gaps.

Next #399 slice: fake bridge metadata/server fixtures for discovery, stale heartbeat handling, subscription normalization, and control request/result matching outside the Desktop remote-control path.

[cbusillo]

Tracker update: #399 advanced again via merged PR #418.

Landed in #418:

• Core fake Code Bridge metadata/server/control fixtures.
• Additive codex_core::code_bridge primitives for metadata discovery, subscription normalization, and control request/result handling.
• Fake WebSocket coverage for auth, subscribe, forwarded control, success, and failure while preserving the Desktop/app-server remoteControl/* boundary.

Validation recorded on #418 / #399:

• focused codex-core lib and integration tests
• git diff --check
• ./build-fast.sh
• GitHub Blob size policy

Next #399 slice remains local navigation/browser/TUI snapshot fixture coverage before any broader port.