release

/release [] — cut a release tag + run the retro

Outer-loop skill. Cuts the release tag and triggers the cross-PR playbook update via the retro skill. Cut ≠ deploy — deploying/arming a release is a separate, human-gated repo-local step (never performed by this skill).

Procedure

1. Target sha. Default = git rev-parse HEAD. MUST be an ancestor of origin/main (release only from merged main).
2. Assert CI is green on that exact sha. At least one check-run exists and every conclusion ∈ {success, neutral, skipped}; any pending / failure / cancelled / timed_out blocks the cut. "Merged" ≠ "shippable" until CI passed on that sha.
3. Show the release window to the user:
   • git log <prev_tag>..<sha> --oneline
   • gh pr list --state merged --base main --search "merged:>=<prev_tag_date>" --limit 60
   • gh issue list --state closed --search "closed:>=<prev_tag_date>" --limit 60
4. Wait for owner OK on the release cut.
5. Cut the tag. Use the repo's release_cmd (.claude/sdlc.toml) if set; otherwise the generic fallback:

   git tag -a <tag> <sha> -m "Release <tag>" && git push origin refs/tags/<tag>
   gh release create <tag> --target <sha> --generate-notes --title <tag>
   

   Tag naming is the repo's choice (e.g. semver v0.1.0 auto-bumped from the latest tag).
6. Invoke /retro --window <prev_tag>..<sha> for the cross-PR synthesis — it spawns a subagent to read the window's merged PRs + closed issues + git, then opens a single playbook PR with higher-order patterns the per-PR review passes wouldn't have caught.
7. Surface the playbook PR URL to the owner.
8. Stop. Deploy / arm / promote are separate human-gated steps (especially when prod_gated = true).

Constraints

• Never auto-deploy / auto-arm / auto-promote.
• The release tag is durable; only cut after owner approval + green CI on the exact sha.
• The playbook-update PR is a normal PR — bot review + qa apply.
• One playbook PR per release.

Relation to review

review is the per-PR skill that fires on every merge — it consumes one asset's ## Report and applies playbook deltas continuously. retro (this skill's downstream) is the per-release skill — it synthesizes across the full window of already-merged workstreams for higher-order patterns no single PR's report would surface alone. By the time /release runs retro, the per-PR review passes have already drained the single-PR lessons; the retro's job is the cross-cutting residue.