feat: add post-PR lifecycle phases and ship method

[stack72]

Summary

• Adds two new phases (pr_failed, releasing) between pr_open and done for visibility into CI failures and release builds
• Adds three new methods: pr_merged (pr_open → releasing), pr_failed (pr_open → pr_failed), ship (releasing → done)
• Adds pr-cooldown check enforcing a 3-minute wait after link_pr before checking PR status, giving CI time to run
• Enables recovery from pr_failed via link_pr (re-link) or implement (major rework)
• Adds skill guidance requiring human confirmation before opening PRs

Test Plan

• deno check — type checking passes
• deno lint — no lint errors
• deno fmt — all files formatted
• deno run test — all 4260 tests pass (26 in lifecycle + schema tests)
• deno run compile — binary compiles successfully
• Manual: create an issue-lifecycle instance, walk through link_pr → pr_merged → ship flow
• Manual: verify pr_failed → link_pr recovery loop works

🤖 Generated with Claude Code

[github-actions]

Code Review

Blocking Issues

1. plan-approved check blocks the pr_failed → implement recovery path (extensions/models/issue_lifecycle.ts:270)
   The TRANSITIONS table now allows implement from pr_failed, but the plan-approved check hard-codes state.phase !== "approved" — so calling implement from pr_failed will always be rejected by the check, even though the valid-transition check passes. The recovery path documented in the skill and implementation reference is broken at runtime.

   Fix: update the plan-approved check to also allow pr_failed as a valid phase, e.g.:

   if (state.phase !== "approved" && state.phase !== "pr_failed") {

Suggestions

1. No tests for the pr-cooldown check — The cooldown check has non-trivial time-based logic (reading linkedAt, computing elapsed time, formatting remaining seconds). A test with a recently-linked PR and a test with an old-enough linkedAt would give confidence this works. The check can be tested via the model's check infrastructure or by calling its execute directly.

2. No test for link_pr from pr_failed recovery path — link_pr now accepts pr_failed as a source phase, and the PullRequestSchema documents that failedAt/failureReason are "cleared on next link_pr." Adding a test that calls link_pr after seeding a pullRequest-main resource with failure fields would verify the recovery path and that those fields are indeed absent in the overwritten resource.

3. link_pr description is stale (issue_lifecycle.ts:1170) — The description says "Transitions the phase to pr_open if currently implementing" but it now also handles recovery from pr_failed. Consider updating to mention both source phases.

4. start transition test doesn't cover new phases (schemas_test.ts:44-47) — The test only asserts pr_open is included in start's allowed phases. The two new phases (pr_failed, releasing) are also in the list. A minor gap — the Phase ordering test catches their existence, but an explicit assertion on start would document the intent.

[github-actions]

Code Review

Well-structured PR that adds post-PR lifecycle phases with clean state machine design.

Blocking Issues

None.

Suggestions

1. as PullRequestData cast in pr-cooldown check (schemas.ts:310-312): The raw JSON.parse + type assertion bypasses Zod validation. If the stored data is malformed (e.g., linkedAt missing), new Date(pr.linkedAt).getTime() would silently produce NaN, and the cooldown check would always pass. Consider parsing through PullRequestSchema.parse() for defense-in-depth — though this follows the existing pattern in the codebase, so not blocking.

2. Context type duplication across methods: The context type for pr_merged, pr_failed, link_pr, and ship is repeated inline. A shared type alias (e.g., MethodContext) could reduce the surface area for drift. Low priority — the current approach is explicit and works.

DDD Assessment

Good domain modeling:

• State transitions via explicit domain methods (pr_merged, pr_failed, ship) rather than generic setters — proper ubiquitous language.
• The pr-cooldown check encapsulates a domain policy as a named concept.
• Recovery paths (pr_failed → link_pr, pr_failed → implement) model real-world scenarios explicitly.
• Attempt tracking on PullRequestData gives the aggregate history awareness without separate event sourcing.

Checklist

• No any types
• All promises awaited (no fire-and-forget)
• License headers present on all .ts files
• No imports from internal libswamp paths
• Test coverage for all new methods and the cooldown check
• Schema tests updated for new phases and transitions
• Upgrade entry added for version bump
• Skill docs updated to reflect new phases and human-confirmation rule