Upstream sync and Codex Lab feature reclassification

Objective

Own the post-MVP program that brings Codex Lab onto current upstream substrate while keeping Every Code product differentiation explicit, thin, testable, and recoverable. This issue is the durable umbrella for convergence execution and Background Review architecture, not a commit ledger.

Finish Line

Codex Lab runs on a current upstream-derived substrate; every surviving Every Code feature layer has an explicit ownership decision and executable contract; broader Background Review architecture is reconciled; and permanent dogfood graduation can proceed without hidden fork drift.

Current Status

State: Superseded by the post-cutover roadmap in #498.

The upstream-first source transition this umbrella coordinated is complete: #126, #428, #305-#308, #429, and #343 are closed, and protected main is the converged candidate at 6f94416cf5f7ccd5a8ed4d0b8ae5ca4c080e6df5 with post-merge CI green.

The remaining work no longer belongs in one convergence umbrella:

Sync Codex Lab with current openai/codex substrate #230 now owns steady-state small upstream refresh operations and is intentionally blocked by the first post-cutover release Publish and install first post-cutover Codex Lab prerelease #499.
Rework durable Background Review around upstream review primitives #231 remains a later Background Review architecture/product-maintenance track related to Launch Codex Lab and establish post-cutover roadmap #498; its unfinished work is not represented as complete.
Launch Codex Lab and establish post-cutover roadmap #498 is the canonical launch and product roadmap, with Publish and install first post-cutover Codex Lab prerelease #499 as the sole concrete Now item and Bound and compress local session rollout storage #490 as Next.

Disposition: close this umbrella as not planned because its pre-cutover roadmap shape is obsolete. Preserve #230 and #231 as independent open work.

Next action: none in this issue.

Blocked by: none.

Last verified: July 29, 2026.

Scope

In: convergence program ownership, Every Code feature classification by contract, upstream snapshot intake, Background Review architecture, product-lane sequencing, and graduation relationships.
Out: detailed subsystem implementation, per-commit historical classification, direct release execution, or unrelated product roadmap work.

Acceptance Criteria

Convergence execution has a durable parent and native dependency graph in Sync Codex Lab with current openai/codex substrate #230.
The compatibility envelope and deterministic audit tooling are preserved as completed evidence.
Define Every Code-owned convergence contracts #126 defines the executable Every Code ownership/contract matrix.
Build upstream snapshot integration and canary branch #428 produces a refreshable upstream-first candidate.
Sync canonical turn, context, history, and recovery substrate #305-Sync TUI, Guardian, and CI substrate #308 complete product-lane reconciliation against that candidate.
Rework durable Background Review around upstream review primitives #231 is thinned onto the converged review/hooks substrate without losing Background Review behavior.
Cut over to converged Codex Lab baseline and primary harness #343 completes protected cutover, rollback/re-entry, quality, and soak gates.
Repository/worktree cleanup does not delete unmerged user work or confuse stale branches with active plan state.

Workstreams

Execute Sync Codex Lab with current openai/codex substrate #230's upstream-first convergence graph.
Rework durable Background Review through Rework durable Background Review around upstream review primitives #231 after the candidate substrate is stable.
Keep validation, plugin lifecycle, provenance, remote-control, and dogfood issues related without duplicating their finish lines here.
Use focused implementation issues for concrete risks such as Stop emitting free-form analytics error subreason #429 rather than expanding historical ledger buckets.

Relationships

Post-MVP successor referenced by closed Codex Lab MVP dogfood plan #28.
Native children: Sync Codex Lab with current openai/codex substrate #230 convergence execution and Rework durable Background Review around upstream review primitives #231 Background Review rework; Classify Codex Lab feature ownership after upstream sync #232/Clean up stale Codex Lab worktrees and planning artifacts after sync classification #233 remain completed history.
Graduation gate: Cut over to converged Codex Lab baseline and primary harness #343.
Related product tracks: Plugin lifecycle hooks for workspace preparation and quality providers #149 plugin lifecycle, Restore Every Code auto-validation loop #284 validation, Make dogfood build provenance visible #311 provenance, and remote-control/dogfood plans.

Validation

Issue graph reflects one active next action and native blockers.
Parent/child Current Status sections agree on strategy, candidate, and cutover ordering.
Project Focus keeps exactly one meaningful Now item.

Decisions

Stop using exhaustive commit classification as the gateway to implementation.
Preserve ledgers as evidence and adopt current upstream snapshots by default outside explicit Every Code contracts.

Open Questions

None at the umbrella level; child issues own implementation questions.

Plan direction checkpoint, 2026-07-04:

After PR #270, upstream alignment is the next active workstream, but agent consensus is to continue the staged strategy rather than do a broad merge.

Fresh refs:

origin/main: 95161fe386810dd2ddb58738fee32eb9715d91ce
openai/main: 98d28aab54ed86714901b6619400598598876dd0
Divergence: 444 / 730 commits across origin/main...openai/main
Hot-path upstream delta since 4e259c3bd0: about 495 commits touching app-server, app-server-protocol, core, login, TUI, MCP/protocol/thread-store surfaces.

Next action:

Use Define Every Code-owned convergence contracts #126 for the upstream drift + ownership ledger before the next implementation PR.
The ledger should classify clusters as reuse/import, thin local layer, Every Code-owned differentiation, defer, or retire/quarantine.
The first implementation slice after the ledger should be a narrow auth/account/thread substrate sync, preserving loaded-thread auth refresh behavior from Refresh loaded threads after API key login #267-Refresh loaded thread auth on logout #270.

Anti-goals:

No 730-commit broad upstream merge as the next PR.
No reimport of upstream TUI ChatgptAuthTokensRefresh stubs over Every Code-owned behavior without an explicit product decision.
No new auth/TUI/thread feature expansion until the overlap with upstream is classified.

Ownership matrix update from #126 drift ledger checkpoint 1:

The first protected local surfaces are now explicit:

loaded-thread auth refresh from Refresh loaded threads after API key login #267-Refresh loaded thread auth on logout #270;
live local account APIs account/list, account/remove, account/switchActive;
third-party agent prompt delivery, CLI preflight, and status UX;
skills/hooks regression hardening;
Code Bridge and durable auto-review as Every Code product surfaces.

Upstream surfaces to import in staged lanes:

auth serialization/storage and account-shape additions;
thread-store/session identity/history substrate;
MCP runtime/catalog/OAuth/retry substrate;
newer agent-control substrate after local third-party guarantees are re-homed;
SkillsService / AGENTS.md / blocking hook substrate after local hook tests are kept green.

Next concrete action is an invariant-guard PR, not the sync itself: add the missing TUI test that account switching stays on the app-server account/switchActive path and does not degrade into session restart semantics.

Stage 2 substrate checkpoint landed: PR #273 (Share resumed rollout history) merged to main as 0635172b711e4fd08deb1cb862c337dac5855c92.

This imports upstream openai/codex@330ae6a516 as a narrow continuation after #272, sharing resumed rollout histories with Arc<Vec<RolloutItem>> while preserving Every Code session identity/provenance behavior. Durable recovery details are in #230.

Stage 2 persistence guardrail landed: PR #274 (Close thread persistence on channel close) merged to main as 4b1c024936ad0fa6112c2a425722bc99eddf268e.

This imports upstream openai/codex@c55ce3b51b, closing live thread persistence when the submission channel closes without an explicit shutdown. Durable recovery details are in #230.

Stage 2 breadcrumb: PR #275 landed the terminal rollout event durability slice.

PR: Flush terminal turn events to rollout #275
Merge commit: b5f1eefe32b0cd30d4c004abd3c30c535f75a6f2
Upstream slice: openai/codex@f5f812389e
Result: TurnComplete and TurnAborted terminal events now flush rollout persistence after append, with focused codex-core tests covering completion/abort flush counts.
CI passed after rerunning an unrelated macOS app build disk-space failure.
Detailed recovery status is in Sync Codex Lab with current openai/codex substrate #230.

Stage 2 substrate breadcrumbs since the last parent-plan update:

PR Avoid duplicate history read on cold resume #276 merged as 561153e1c8, avoiding duplicate history reads on cold resume while preserving Every Code session/provenance behavior.
PR Add rollout turn item variants #277 merged as ec2ee9a33a, adding missing rollout TurnItem variants plus app-server v2 conversion/schema/replay support and downstream analytics/TUI exhaustiveness handling.

Durable recovery details remain in #230. Current direction is still staged upstream substrate sync, not a broad fast-forward. The next likely slice is ThreadHistoryMode / history-mode persistence, with initial context metadata and descendant listing behind it.

Navigation Menu

Repository navigation

Upstream sync and Codex Lab feature reclassification #229

Objective

Finish Line

Current Status

Scope

Acceptance Criteria

Workstreams

Relationships

Validation

Decisions

Open Questions

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

Upstream sync and Codex Lab feature reclassification #229

Description

Objective

Finish Line

Current Status

Scope

Acceptance Criteria

Workstreams

Relationships

Validation

Decisions

Open Questions

Activity

shiny-code-bot commented on Jul 4, 2026

shiny-code-bot commented on Jul 4, 2026

shiny-code-bot commented on Jul 5, 2026

shiny-code-bot commented on Jul 5, 2026

cbusillo commented on Jul 5, 2026

cbusillo commented on Jul 6, 2026

Metadata

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

Issue actions