Publish and prove Codex Lab lab.5

[shiny-code-bot]

Objective

Publish and prove codex-lab-v0.1.0-lab.5 as an external-agent reliability release containing three coherent repairs: canonical Gemini/Antigravity routing, authoritative Claude quota diagnostics, and deterministic typed local-agent behavior failures. Keep unrelated product features and broad provider-value evaluation outside the release.

Finish Line

Another developer can upgrade from lab.4 to signed lab.5 and prove all three reliability lanes: Gemini launches through the normal configured agent.create route with a canonical advertised selector and non-empty provenance-bound result; Claude reports provider-supplied quota window, reset, and overage state without contradictory prose; local-agent behavior scenarios terminate with deterministic typed outcomes and bounded tool-loop handling. Restart/resume, rollback to lab.4, restoration to lab.5, and one real desktop update/relaunch also pass.

Current Status

• The latest published prerelease remains codex-lab-v0.1.0-lab.4; no lab.5 tag or release exists as of August 19, 2026.
• Authoritative validation-only run 32201700621 completed from exact main@39d2eac3bfaaf32d52b49fda503b1aceabec3aed with 55 hard Rust failures and all non-Rust release infrastructure green.
• PRs Correct executor skill read cursor contract #742–Stabilize curated plugin auth-switch coverage #750 have since merged, most recently curated plugin auth-switch coverage repair Stabilize curated plugin auth-switch coverage #750 as 2a8e00f9c16e1935f3bc1fc093cfde0474d202e3. Together they are expected to clear twelve failures, leaving approximately 43; this remains an estimate pending later exact-SHA remeasurement.
• All nine repair PRs passed 17/17 CI and required Opus plus Gemini/Antigravity review. Publication remains prohibited: continue bounded convergence repairs, then require a complete green validation-only run from one frozen SHA before publishing from that same SHA.

Acceptance Criteria

• Canonical Gemini/Antigravity selector and normal-route proof are implemented.
• Claude quota diagnostics report provider-supplied window, reset, and overage state.
• Local-agent behavior failures terminate with deterministic typed outcomes and bounded tool-loop handling.
• Full lab.5 release identity and exact-provenance verification are implemented across runtime and packaging surfaces.
• Publish signed codex-lab-v0.1.0-lab.5 from an exact green main commit.
• Upgrade a real installation from lab.4 to lab.5 and prove all three reliability lanes after restart/resume.
• Roll back to lab.4, restore lab.5, and verify state/provenance continuity.
• Complete one real desktop update/relaunch acceptance proof.

Scope

• In: Reject malformed discovered external-agent selectors before launch #581 Gemini/Antigravity selector normalization or suppression and live normal-route proof; Surface structured rate-limit diagnostics for Claude CLI agents #315 Claude structured quota diagnostics; Make local-agent behavior tests deterministic and failure-typed #620 typed deterministic local-agent failures and bounded tool-loop handling; exact-provenance lab.5 publication; lab.4 upgrade/rollback; Harden websocket reconnect after rapid desktop relaunch #464 real desktop update/relaunch acceptance.
• Out: Broad marginal-value evaluation Measure review validation and agent marginal value #523; new provider integrations; automatic provider substitution; general approval-friction work Reduce managed approval friction in dogfood workflows #313; unrelated review/validation UX; broad local-model selection policy; billing or quota-management changes.

Relationships