This repository was archived by the owner on Oct 9, 2026. It is now read-only.
Repository navigation
This repository was archived by the owner on Oct 9, 2026. It is now read-only.
MVP classification ledger for restored Every Code evidence #30
Description
Objective
Build the evidence ledger that keeps restored cbusillo/code history useful without turning it into an automatic port backlog. Each candidate behavior should be classified by current Codex Lab value and target substrate.
Finish Line
Every MVP candidate from restored Every Code evidence is classified with rationale, target surface, and validation gate.
Current Status
State: Ledger drafted from restored Every Code evidence and current Codex Lab proof. The MVP spine is narrower than the old Every Code surface: protect local CLI dogfooding first, keep Code Bridge as a standalone service/protocol track, and defer discord-blue/Launchplane cutover until remote-session continuity has proof.
Source anchors used:
- Restored Every Code inventory:
/Users/cbusillo/Developer/code/docs/codex-base-feature-inventory.md - Restored fork parity ledger:
/Users/cbusillo/Developer/code/docs/codex-fork-parity-ledger.md - Restored Every Code README:
/Users/cbusillo/Developer/code/README.md - Current Codex Lab harness evidence:
tools/codex-exec-harness/README.md,tools/codex-exec-harness/scenarios/skills-cache-continuity.json,tools/codex-exec-harness/harness.py - Current app-server gate:
codex-rs/app-server/README.md - Parent plan and decisions: Codex Lab MVP dogfood plan #28, Scope minimal Code Bridge vertical slice #36, Rescope Auto Drive for Codex Lab MVP #37, Define Code Bridge protocol, trust, and payload contract #45, Build minimal Code Bridge service skeleton #46, Add Code Bridge app client and subscriber vertical slice #47, Align discord-blue Every Code bridge with shared protocol #48, Plan Launchplane provenance hooks for Code Bridge #49, Gate thin app-server integration for Code Bridge #50
| MVP candidate | Classification | Rationale | Target / evidence / old surface | Validation gate |
|---|---|---|---|---|
| Repository identity and durable planning recovery | Covered | The active fork/source-material split is settled, and future recovery points live in GitHub plan issues. | Evidence: #28 plus repo AGENTS.md points agents at cbusillo/codex-lab as active and cbusillo/code as restored history only. |
No code gate; keep Current Status in plan issues current. |
| Exec-harness reliability substrate | Covered | The harness is now the MVP regression spine for agent behavior, cache expectations, and proof-loop checks. | Evidence: #31 / PR #38 and tools/codex-exec-harness/README.md. |
just exec-harness-test. |
| Skills, project-doc, and thread cache continuity | Covered | The two-turn scenario proves resumed thread continuity without duplicate prompt bloat. | Evidence: #33 / PR #40 and tools/codex-exec-harness/scenarios/skills-cache-continuity.json. |
just exec-harness-test; scenario must assert same thread, marker counts, cached token floor, and cache ratio. |
| Prompt-cache and token usage assertions | Covered | The MVP needs cache regression protection, and the harness already exposes deterministic usage/cached-token checks. | Evidence: #32 / PR #39 and harness usage assertion docs. | Harness assertions over input tokens, cached input tokens, cache ratio, prompt bounds, and prefix stability. |
| Auto Review proof loop | Covered | The useful piece is bounded, actionable proof of current/stale/detached review findings; the old background watcher does not need wholesale restoration. | Evidence: #35 / PR #44 and #51; harness.py bounded summary/detail helpers and run sidecars. |
Exec-harness unit tests for current/stale/detached findings, clean-run quietness, summary truncation notices, and stable detail lookup. |
| Codex Desktop / app-server compatibility | Covered | App-server v2 is the compatibility substrate; Every Code/Codex Lab extensions must be additive. | Evidence: #34 / PR #41 and codex-rs/app-server/README.md. |
just write-app-server-schema, just test -p codex-app-server-protocol, and targeted app-server/CLI tests before protocol overlays. |
| App-server RemoteControl transport | Covered | RemoteControl pairing/enrollment is present upstream and is not the same thing as Code Bridge. | Evidence: app-server RemoteControl docs and protocol types. | Existing Desktop compatibility gates must stay green when bridge-adjacent work lands. |
| Code Bridge app telemetry/control | Rewrite | Do not bury Code Bridge inside app-server. Rebuild it as a standalone service/daemon plus shared protocol, with app-server only as a thin integration gate if needed. | Target: #45-#50; service owns console/error/pageview/screenshot/control events, local auth/trust, payload caps, and subscriber/control APIs. | Protocol contract tests; loopback service tests; sample app emits telemetry and accepts screenshot/control; app-server schema fixtures unchanged except intentional thin gate. |
| discord-blue external-session bridge | Later | discord-blue is a product surface, but it depends on proven Codex Lab remote-session lifecycle rather than Code Bridge itself. It should not drive the first local CLI dogfood cutover. |
Target: after remote-session proof, validate the existing product bridge against Codex Lab endpoints or update both sides deliberately. | End-to-end external session discovery, resume/continue, approval/reply, heartbeat/reconnect, and audit trail. |
| Launchplane work-request / provenance hooks | Later | Launchplane should integrate through explicit work-request/provenance contracts after the local dogfood spine and bridge protocol are stable. | Target: #49 and later Launchplane round-trip issue; keep Launchplane as orchestration/provenance, not Code Bridge authority. | Redacted provenance contract tests plus a live work-request to Codex Lab session pickup round trip. |
| Broader token/rate-limit diagnostics UI | Later | Harness-level cache protection is MVP; richer TUI/app-server diagnostics can wait until the dogfood baseline shows real gaps. | Target: current token usage events and app-server/TUI surfaces, not old Every Code UI wholesale. | Audit current events first; add focused snapshot/protocol tests only for proven missing diagnostics. |
| Agent selector / multi-agent UX parity | Later | Agents are useful now through current tooling, but old selector/status UI parity is not a cutover blocker. | Target: Codex-native subagent/model-provider primitives if dogfooding shows friction. | Fixture coverage for selector resolution, grouped agent status, error propagation, and local model config before UI work. |
| TUI settings, history, and polish overlays | Later | Old Every Code UI polish should not expand the MVP unless a live dogfood workflow needs it. | Target: focused Codex TUI overlays only. | just test -p codex-tui plus cargo insta pending-snapshots -p codex-tui / reviewed snapshots for intentional UI changes. |
| Docs and release promises | Rewrite | Docs must reflect the narrower Codex Lab MVP and avoid promising restored Every Code parity ahead of shipped gates. | Target: Codex Lab README/release notes/planning issue prose, not general product docs in docs/. |
Docs diff with no unbuilt parity claims; link to the owning feature gate or issue for every promised surface. |
Auto Drive coordinator, cards, settings, and /auto flows |
Retire | Codex goal/thread/session primitives cover the useful semantics closely enough for MVP; old coordinator state would duplicate product direction. | Old surfaces not returning: Auto Drive coordinator crates, TUI cards, settings pane, /auto, code exec --auto, countdown/observer/review toggles, Auto Drive-specific docs. |
None; any future unique gap needs a new issue with a Codex-native validation gate. |
| App-server v1 conversation schema resurrection | Retire | The MVP should not restore old v1 conversation/root event models absent a specific external contract. | Old surfaces not returning: v1 conversation schema resurrection and root EventMsg/old TS request-event models. |
None; future external client needs must be expressed as additive v2/experimental protocol work. |
Decision summary:
- Green for local dogfooding spine: exec harness, cache/thread continuity, Auto Review proof loop, and app-server compatibility are covered.
- Green to plan Code Bridge as a service: yes, Code Bridge should run as its own local service/daemon with a shared protocol; app-server should only get a narrow integration gate if the protocol needs Desktop/session awareness.
- Not green to merge Code Bridge, discord-blue, and Launchplane into one feature: those are adjacent integration tracks with different authorities and validation gates.
Recommended next action:
Run a dogfood readiness sweep over the covered spine and stale docs, then start #45/#46 only after the cutover baseline is clean. Code Bridge should remain scoped by #36 and its subissues, not folded into app-server wholesale.
Scope
In:
- Classify restored source material as
Port,Rewrite,Covered,Defer, orRetire. - Link evidence from
cbusillo/codeissues, commits, docs, tests, or fixtures. - Name the Codex Lab target surface for any candidate that remains in MVP scope.
Out:
- No wholesale module copying from
cbusillo/code. - No implementation work.
- No app-server v1 resurrection unless a specific client contract is proven.
Acceptance Criteria
- Ledger includes the source anchors named in Codex Lab MVP dogfood plan #28.
- Each item has exactly one classification and a brief rationale.
- Items marked
PortorRewritename the intended Codex-native surface and validation gate. - Items marked
Coveredlink current Codex Lab evidence. - Items marked
Retirestate which old surface should not return.
Relationships
Child of #28. Related to #1 for the broader fork overlay capability map.
Validation
Review the ledger against #28 and restored cbusillo/code evidence before opening implementation PRs for feature ports.
Decisions
Restored Every Code history is evidence, not an implementation base. Port means preserve capability, not code shape.
Open Questions
Resolved. The restored evidence has enough signal to classify the MVP candidates now. Remaining uncertainty is no longer classification work; it belongs in follow-up implementation issues, primarily #36 and its Code Bridge service subissues.
Activity
shiny-code-bot commentedon Jun 12, 2026
Completed. The ledger now classifies restored Every Code evidence against the Codex Lab MVP and records the key boundary decisions: