Launch Codex Lab and establish post-cutover roadmap

[shiny-code-bot]

Objective

Own the post-cutover Codex Lab roadmap after the MVP and upstream-first source cutover. Keep launch, reliability, product value, platform bets, and continuous upstream sync separate so only one concrete item is active at a time.

Finish Line

A signed, provenance-verified Codex Lab private release is installed through the supported path and used successfully by a second developer; known limitations are explicit, rollback is proven, and subsequent product work is sequenced from real dogfood evidence.

Current Status

State: parked roadmap parent while the scoped lab.5 release and acceptance work executes through #624.

The current release hierarchy is #624 → #703 → #707. PR #716 merged on August 17, 2026 at main@780e5f32effbaeacf8a5dccdb0728b5a7bed5d1b; no lab.5 tag or release exists. #707 is the sole active Project Now lane with 16 measurable MultiAgentV2 residual failures. #706 is waiting with its original portability clusters, #708 is Next, #709 is blocked by #708, and #710-#713 remain later.

#464 remains waiting for a published and installed lab.5 artifact so it can supply the real desktop update/relaunch proof before #624 closes. Post-release upstream convergence #230 and prevention work #664 remain parked until #624 completes.

Consumer-backed structured inference issue #319 is a native sub-issue, blocks cbusillo/mediaforce#176, and its implementation merged through PR #719 at main@9e3eeb23e39d63e2e8c4328d515ce33d5c5213be. It remains a waiting platform lane outside the current lab.5 critical path unless #624 explicitly adopts the merge; its next gate is inclusion in an official release followed by downstream Mediaforce acceptance.

Next action: classify and land the smallest coherent #707 product-contract stage from the release workstream's current immutable base, then continue #703's bounded lanes until one exact main SHA passes the full validation-only release workflow. Only then publish/install lab.5, run #464 acceptance, and resume the broader roadmap. The merged #319 RPC must not displace #707 from the sole active Now lane unless the release plan explicitly changes.

Last verified: August 17, 2026 after PR #719 merged.

Roadmap

Phase 0 — Private Prerelease

• Publish a signed prerelease from the exact current main commit through the repository release workflow.
• Install it through the supported installer/supervisor path rather than the rebuild-on-launch development shim.
• Verify CLI/TUI launch, live provenance, official Mac controller connectivity, rollback, and release artifact identity.

Phase 1 — Reliability From Real Use

• Bound and compress inactive session rollout storage through Bound and compress local session rollout storage #490.
• Finish only the Prove persistent Codex Lab engine behind official Mac and iOS clients #385/Harden websocket reconnect after rapid desktop relaunch #464 reliability checks that require real post-cutover use.
• Fix poisoned pooled-account eligibility in Expired pooled accounts poison a healthy active Codex Lab login #519 before broader terminal-auth fallback in Switch execution accounts only after terminal authentication failure #599.
• Resume Prove stable native remote control with separate execution auth #381, Auto-recover stale iOS remote task streams without daemon restart #388, or Reduce managed approval friction in dogfood workflows #313 only when the behavior is reproducible or blocks normal work.

Phase 2 — Visible Product Value

• Prioritize Surface per-account Codex reset credits and reset recommendation #383 per-account limits/reset recommendation.
• Preserve explicit quota-window priming in Add explicit opt-in quota-window priming for enrolled accounts #600 as deferred follow-up after the read-only evidence is useful.
• Complete Adopt Background Review terminology across product surfaces #303 user-facing Background Review terminology.
• Select Plugin lifecycle hooks for workspace preparation and quality providers #149 plugin lifecycle or Rework durable Background Review around upstream review primitives #231 Background Review thinning only after the release and reliability work are stable.

Phase 3 — Platform Bets

• Keep Expose a generic Work Pulse activity stream #298 Work Pulse, Explore cache-aware intent checkpoints for agent drift #150 intent checkpoints, Surface structured rate-limit diagnostics for Claude CLI agents #315 provider diagnostics, and similarly broad architecture work parked until dogfood supplies a concrete need and sequence.
• Do not add an /auto coordinator or duplicate GitHub-backed planning in thread-local goal state without new evidence.

Always-On — Upstream Sync

• Reframe Sync Codex Lab with current openai/codex substrate #230 as steady-state small upstream refresh operations with bounded lag and independently reviewable PRs.
• Do not mix upstream maintenance with the single active launch/product item.

Acceptance Criteria

• A signed private prerelease is published from an exact merged main commit and installed through the supported path by a second developer.
• The installed runtime reports exact release/source provenance and does not rebuild from source during normal launch.
• One bounded second-user smoke proves install, launch, a real code-changing turn, one external-agent operation, one Automatic Validation result, one visible Background Review outcome, restart/resume, and rollback.
• Authentication, permissions, model/provider selection, core status surfaces, known limitations, failure reporting, and rollback are documented in a short recipient guide.
• Scope enters the private-dogfood milestone only when the second-user canary cannot complete install, launch, the core harness smoke, or rollback without it.
• Post-release priorities are selected from observed dogfood evidence rather than open-issue count or architectural adjacency.

Relationships

• Successor to closed Codex Lab MVP dogfood plan #28 MVP planning and closed Cut over to converged Codex Lab baseline and primary harness #343 protected cutover.
• Continuous upstream maintenance: Sync Codex Lab with current openai/codex substrate #230.
• Reliability: Bound and compress local session rollout storage #490, Prove persistent Codex Lab engine behind official Mac and iOS clients #385, and Expired pooled accounts poison a healthy active Codex Lab login #519, with waiting follow-ups Prove stable native remote control with separate execution auth #381, Auto-recover stale iOS remote task streams without daemon restart #388, Harden websocket reconnect after rapid desktop relaunch #464, Switch execution accounts only after terminal authentication failure #599, and Reduce managed approval friction in dogfood workflows #313.
• Product value: Surface per-account Codex reset credits and reset recommendation #383, Add explicit opt-in quota-window priming for enrolled accounts #600, Adopt Background Review terminology across product surfaces #303, Plugin lifecycle hooks for workspace preparation and quality providers #149, and Rework durable Background Review around upstream review primitives #231.
• Long-horizon platform work: Expose a generic Work Pulse activity stream #298, Explore cache-aware intent checkpoints for agent drift #150, and Surface structured rate-limit diagnostics for Claude CLI agents #315.
• Closed product-direction decision: Design durable goals as the substrate for auto execution #121; GitHub planning remains canonical and no /auto investment is planned.
• related: Adopt Background Review terminology across product surfaces #303 - Adopt Background Review terminology across product surfaces #303
• related: Rework durable Background Review around upstream review primitives #231 - Rework durable Background Review around upstream review primitives #231
• related: Sync Codex Lab with current openai/codex substrate #230 - Sync Codex Lab with current openai/codex substrate #230
• related: Prove persistent Codex Lab engine behind official Mac and iOS clients #385 - Prove persistent Codex Lab engine behind official Mac and iOS clients #385

Decisions

• Ship privately before considering a public beta.
• Replace the development launcher experience with a signed installed release before adding product scope.
• Reliability and storage follow the release; feature expansion does not block launch.
• Keep upstream sync continuous and small, but separate from the product roadmap.
• Use GitHub issues and the native issue graph as the canonical durable planning state for GitHub-backed work.
• Do not build an /auto coordinator or invest in /goal as a second planning system without new concrete evidence; the existing /goal surface may remain optional for local/non-GitHub compatibility.
• Keep poisoned-account isolation (Expired pooled accounts poison a healthy active Codex Lab login #519), terminal-auth execution fallback (Switch execution accounts only after terminal authentication failure #599), and explicit quota-window priming (Add explicit opt-in quota-window priming for enrolled accounts #600) as separate product and implementation boundaries.

Open Questions

• Select the prerelease version/tag in the release child.
• Decide whether public beta readiness deserves a later dedicated phase after private dogfood evidence.