ci(pr-automation): reconcile lifecycle state

[Benoît Cortier (CBenoit)]

Derive next-actor labels from current PR, CI, lease, and trusted exact-head review state instead of individual event transitions.

Preserve already-admitted valid reviews across draft and closed-unmerged states while stale, merged, ambiguous, and partial states fail closed.

[Copilot]

Copilot review overview

🟡 Changes recommended

Removing the oversized-review authorization no longer stops an active run from using its stale elevated evidence limit.

Review effort: Balanced
Findings: 1

Open (1)

• Revoking oversized authorization does not cancel running classification · New

What changed in this PR

Centralizes PR lifecycle reconciliation around live PR, CI, lease, and review-receipt state.

Changes:

• Adds lifecycle reconciliation and trusted review outcome receipts.
• Serializes PR mutations while retaining queued wake-ups.
• Strengthens review admission and closed/draft publication guards.

File	Description
.github/​workflows/​pr-automation.yml	Adds lifecycle events, serialization, and reconciliation.
.github/​workflows/​pr-automation.intent.md	Defines the revised lifecycle contract.
.github/​pr-automation/​write-state.js	Guards and sequences result publication.
.github/​pr-automation/​review-outcome.js	Encodes trusted review receipts.
.github/​pr-automation/​resolve-state.js	Delegates actor labels to reconciliation.
.github/​pr-automation/​resolve-pr.js	Preserves PR identity for lifecycle-only runs.
.github/​pr-automation/​lifecycle.js	Derives and writes actor labels.
.github/​pr-automation/​current-head.js	Supports guarded closed-unmerged continuation.
.github/​pr-automation/​ci-state.js	Orders same-generation CI states.
.github/​pr-automation/​automation.test.js	Expands lifecycle and publication coverage.
.github/​pr-automation/​automation-lease.js	Revalidates admission and publication leases.
.github/​PR_AUTOMATION.md	Documents lifecycle behavior and safeguards.

[Benoît Cortier (CBenoit)]

PR automation is failing because of a picky-krb 0.12.5 incompatibility, fixed on master by #2074. Please rebase on master to fix it.

Update: no rebase needed anymore. picky-krb 0.12.5 was yanked from crates.io (re-released as 0.13.0), so the API check builds again without changes to this branch. PR automation has been re-run here and passes.

[github-actions]

PR 2068 replaces per-event actor-label deltas with snapshot-derived lifecycle reconciliation, removes the ai-review/allow-oversized machinery, and serializes admission, publication, and reconciliation through one per-PR mutation group. The change is a net deletion with careful fail-closed guards and strong test coverage; no correctness defects were found. Remaining review value is compression: the new guard closures in write-state.js duplicate one skeleton with a redundant ownership check, lifecycle.js re-reads a full snapshot per removed label, and small helpers (classification guard, closed-unmerged predicate, label-name mapping, unused export) duplicate existing ones. All six specialist findings were independently verified against the head and accepted as low-risk, behavior-preserving consolidations.

1. [code-compressor] isOpenNonDraftAtHead is exported and imported but never called outside current-head.js — low 🟡 — .github/pr-automation/current-head.js
   The pure helper exists only as the body of isOpenNonDraftAtHeadNow, yet module.exports lists it and write-state.js imports it while using only isOpenNonDraftAtHeadNow (line 393); automation-lease.js likewise uses only the Now variant. Dropping the export and the unused import keeps the helper private with zero behavior change. Line range is null because the export statement is a modified rather than fully added line.

[github-actions]

The PR reworks PR automation so next-actor labels are derived by lifecycle reconciliation from live PR, CI, lease, and trusted review-receipt state instead of per-event transitions, adds versioned review-outcome receipts, extends eligibility to closed-unmerged continuation, and removes the ai-review/allow-oversized feature. The change is large but internally consistent, fail-closed, and covered by focused tests; the author's review responses defend the phase-specific guards plausibly. The validated code-compressor candidates are all low-severity maintainability items: one unused export plus a dead import, a duplicated ACTOR_LABELS constant, and two verbatim guard duplications in write-state.js. Three are accepted as-is; the review-gate duplication candidate is refined because the two closures are not identical apart from the phase string: canPublishReviewReceipt performs an extra canMutate precondition that canCleanupReview deliberately omits, so the proposed factory must parameterize…