| name | release-preparation |
| description | Use when running /release:release — gathers completed features, checks release readiness, and drafts changelog/release notes. Never performs the release itself (no git tag/push/deploy). Philosophy adapted from BMAD-METHOD's governance discipline and this toolkit's evidence-before-claims principle. |
Release Preparation
This skill exists to answer one question honestly: "is this actually ready to release, and
what changed?" — then draft the documentation, never execute the release itself. Performing
git tag, git push, or a deploy is explicitly out of scope (see
.claude/CONSTITUTION.md principle 7); this skill's output is always a draft for a human to
act on.
Readiness gate (all must hold before drafting anything)
- Tests are green — fresh. Run the full suite now; do not reuse a result from earlier in
the conversation. This is the
verification-before-completion discipline applied to the
whole release, not one task.
- No open
[NEEDS CLARIFICATION] markers in any in-scope feature's Specification.md.
- Every in-scope feature's
Tasks.md is fully checked off. A partially-done feature is
not release material unless the user explicitly says otherwise.
If any item fails, stop and list it as a blocker — do not draft the changelog or PR
description around an unmet gate. A caught blocker is the whole point of this skill.
Process (once the gate passes)
- Diff against the last changelog entry to find what's genuinely new.
- Summarize each completed feature in one line, sourced from its
Specification.md
overview — not from guessing at what the code does.
- Draft the changelog entry and a PR/release description.
- List the manual commands a human would run next (tag, push, publish, deploy) — clearly
labeled as suggestions, never executed by this skill.
Red flags
| Red flag | Why it matters |
|---|
| Drafting release notes without running tests fresh in this turn | Violates evidence-before-claims — a stale "tests passed" claim is not evidence |
A feature is included whose Tasks.md still has unchecked items | Ships something that was never verified as actually done |
The skill (or the agent using it) runs git push/git tag/a deploy command | Out of bounds — this skill drafts, a human executes |
