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.
Classify Codex Lab feature ownership after upstream sync #232
Description
Finish Line
Third-party agents, auth/account UX, plugins/skills/hooks, Code Bridge, and related Every Code features have a decision matrix that names the upstream substrate to reuse, the local layer to keep, and any code to remove or convert to plugin/skill form.
Current Status
July 1, 2026: ownership matrix drafted from current origin/main (961cd2b before PR #236), current openai/main (db887d0), Every Code reference /Users/cbusillo/Developer/code-prealign-new-skills/code-rs, and Codex Lab issue context. Two read-only agents contributed rows; they could not query cbusillo/code from their sandbox, so future passes should still cross-check that repo directly when scoping PRs.
| Cluster | Every Code source of truth | Upstream surface to reuse | Current Codex Lab surface | Decision | Product risk if removed | Code risk if kept | Smallest next PR |
|---|---|---|---|---|---|---|---|
| Third-party / external agents | Current Codex Lab agent runtime plus Every Code role/config migration ideas in code-prealign-new-skills |
Upstream multi-agent/session/thread lifecycle concepts, not launcher implementations | codex-rs/core/src/agent/*, especially external command backend and role resolver |
Every Code differentiator; align lifecycle to upstream, keep third-party launchers behind role/backend config | High: loses Antigravity/Copilot/Claude/local agent delegation | Medium-high: process spawning, cwd/env isolation, cancellation, timeout, stdout/stderr caps | PR to document/test external-command contract against upstream-style spawn/cancel/timeout/status finalization |
| Agent role packs / built-ins | Codex Lab built-in role TOML + resolver; prealign has migration/import ideas | Upstream role/session metadata where useful | agent/builtins/*.toml, agent_resolver.rs, role.rs |
Thin local layer; move non-essential built-ins toward plugin-packaged role packs | Medium: weakens out-of-box multi-agent UX | Medium: hardcoded roles can drift from plugin/marketplace policy | PR to define agent role-pack manifest and convert one non-critical built-in to a plugin-delivered role |
| Plugins / marketplace | Codex Lab core-plugins |
Upstream plugin manifest/store/marketplace/schema pieces where current and compatible | codex-rs/core-plugins/*, config plugins, app-server plugin schemas, analytics plugin state |
Keep as packaging substrate; avoid broad remote/share policy until product need is explicit | High: blocks packaging skills/hooks/agent packs | High: install/share/auth/archive path is security-sensitive | PR to stabilize local plugin manifest + store + enabled state only |
| Skills | Prealign core/src/skills/* and app-server skills list flow; current Codex Lab split into crates |
Upstream skill metadata/list/remote/read/write protocol concepts and bounded injection tests | codex-rs/core-skills/*, codex-rs/ext/skills/*, available skills context, app-server skills schemas |
Keep and align; primary conversion target for repo/user workflows | High: loses implicitly invokable local expertise and command policies | Medium-high: model-context injection and command policies must stay bounded/safe | PR to reconcile current core-skills with prealign metadata/command-policy behavior |
| Hooks | Current Codex Lab richer lifecycle hooks; prealign simpler project hooks | Upstream hook schema/list/event notification surfaces | codex-rs/config/src/hook_config.rs, codex-rs/core/src/hook_runtime.rs, codex-hooks |
Keep as guarded automation surface; plugins may contribute hooks, trust policy stays central | High: loses lifecycle automation and permission interception | High: hooks execute code; trust hashes and managed-only mode are sensitive | PR for plugin/system/user hook precedence and managed-hooks-only integration coverage |
| Auth/account UX | Prealign account_switching.rs, account_usage.rs, auth_accounts.rs, TUI login/account settings views |
Upstream v2 account API: login start/cancel, API key, device-code/AuthTokens, account/read, rate limits, token usage, notifications | Codex Lab account processor plus Every Code account switching/usage modules | Keep product UX; align auth basics to upstream and isolate Every Code switching/quota continuity | High: loses multi-account workflow and quota continuity | Medium-high: auth touches config, session retry, app-server, token/rate-limit persistence | PR to normalize account processor against upstream and isolate Every Code switching/usage modules with focused tests |
| Durable review / auto-review | Prealign auto-review docs/review coordination plus current Codex Lab durable store/scheduler | Upstream manual review/start, ReviewTarget, session/task review primitives, Guardian isolation concepts |
codex-rs/auto-review, background_auto_review.rs, auto_review_awareness.rs, app-server summary/detail/control APIs |
Every Code differentiator as thin durable scheduler/artifact layer on upstream review | Very high: loses persistence, dedupe, stable run/finding lookup, background continuity, bounded awareness | High: broad cross-cutting runtime can grow stale or unbounded | PR #236 handles rendered-scope skip; next PR should focus on duplicate/control UX, bounded detail lookup, or TUI status polish |
| Code Bridge / browser workflows | Current Codex Lab Code Bridge plus prealign browser crate/tooling | Upstream UI/request-processor conventions where useful; upstream does not own Code Bridge protocol | code-bridge-protocol, app-server bridge processors, core code_bridge handler, browser crate/tools |
Every Code differentiator; Browser is CDP automation, Code Bridge is app telemetry/control; share UX language but do not collapse prematurely | High: loses app debugging loop, screenshot/control, console/error bridge | Medium-high: security-sensitive and model-visible; reserved/partial states can confuse users | PR to finish smallest bridge slice: status + screenshot/javascript control with bounded outputs and protocol tests |
| Plugin-packaged skills/hooks | Codex Lab plugin roots and HookSource::Plugin; prealign treats skills as first-class |
Upstream plugin detail schema concepts listing skills/hooks/apps/MCP servers | plugin skill roots, plugin manifest/detail, hook source, generated TS | Convergence path: plugins package skills, hooks, MCP config, apps, later agent role packs | Medium-high: marketplace value stays thin without packaging | Medium: config precedence and policy can tangle | PR to convert one simple skill bundle into a local plugin fixture and verify list/injection/hook summaries |
Plugin/skill conversion candidates:
- External agent role packs: Antigravity, GitHub Copilot, Claude/Codex-compatible CLIs, qwen, local LM Studio.
- Workflow skills:
github-plan,repo-readiness,babysit-pr,browser-ui-review,local-llm,jetbrains-inspection,work-closeout. - Code Bridge install/setup should be a skill first, plugin later when packaged commands/templates are useful.
- Account switching policy should remain core, not skill/plugin.
- Durable review orchestration can use skills for babysitting and reporting, but durable run storage/scheduling should remain core.
Acceptance Criteria
- A matrix exists for each feature cluster: upstream substrate, Every Code reference files, current Codex Lab files/PRs, product gap, decision, and next implementation slice.
- Each cluster is marked
reuse upstream,thin local layer,Every Code differentiator,convert to plugin/skill, orremove/revert. - The matrix explicitly names what would soften the product and must not be removed merely for upstream alignment.
- Candidate plugin/skill conversions are identified, especially quality providers, third-party agent adapters, browser/Code Bridge workflows, and workspace prep.
- Codex Lab multi-auth MVP #115 is updated or linked with the auth/account conclusion instead of carrying stale backend assumptions.
- Plugin lifecycle hooks for workspace preparation and quality providers #149 is updated or linked with the plugin/quality lifecycle conclusion instead of duplicating hook work.
- The next implementation PR can be scoped from the matrix without rediscovering the upstream/Every Code tradeoffs.
Suggested Matrix Columns
- Cluster
- Every Code source of truth
- Upstream surface to reuse
- Current Codex Lab surface
- Decision
- Product risk if removed
- Code risk if kept
- Smallest next PR
Relationships
- Child of Upstream sync and Codex Lab feature reclassification #229.
- Related to Codex Lab multi-auth MVP #115 multi-auth MVP.
- Related to Plugin lifecycle hooks for workspace preparation and quality providers #149 plugin lifecycle hooks.
- Related to Rework durable Background Review around upstream review primitives #231 for the review row.
Activity
cbusillo commentedon Jul 1, 2026
Completed by the July 1, 2026 ownership matrix now recorded in Current Status.
Evidence:
The matrix remains the durable recovery point for feature ownership decisions under parent #229.