Harden and roll out full merge train

[cbusillo]

Objective

Harden and roll out the full merge train safely across opt-in repositories. This work turns the implementation into an operationally supportable feature with fail-closed guards, cleanup, docs, and smoke evidence.

Finish Line

Harden and roll out full merge train

Current Status

State: Planned. cbusillo/codex-skills is the intended rollout/test-bed repository for the first multi-PR dry-run and low-risk smoke because its changes are isolated.
Next action: Define rollout stages after scheduler and validation strategy are implemented.
Blocked by: All implementation tracks; candidate codex-skills PRs with isolated changes; runner capacity.
Last verified: 2026-05-13; cbusillo/codex-skills:main is configured for the current sequential dry-run policy.

Scope

• In: docs, operator runbooks, cleanup jobs, stale candidate handling, metrics/logging, issue/PR smoke checklist, and activation policy.
• Out: broad automatic enablement for all repos.

Acceptance Criteria

• Docs explain how to opt in/out and how to interpret train records and feedback.
• Cleanup handles stale train refs, cancelled PRs, closed PRs, deleted branches, and superseded policy records.
• Dry-run evidence covers at least two queued codex-skills PRs.
• Mutate smoke merges or safely blocks low-risk codex-skills PRs in order.
• Operators can disable scheduled mutate without code changes.
• Post-merge checks and service health are verified after rollout.
• The codex-skills repo's launchplane skill is updated after the full batch train workflow is implemented and proven.

Relationships

• Blocked by semantics, queue records, speculative validation, scheduler, and feedback work.

Validation

• Full local unit/integration suite.
• CI/Security/CodeQL.
• Deploy Launchplane.
• Live multi-PR dry-run and low-risk mutate smoke.

Decisions

• Activation should remain explicit per repository/base branch through Launchplane-owned records and policy.

Open Questions

• Which repo should be the first full-train mutate smoke after the dry-run proves stable?