agentskills.codes
RE

release

by mattstylesSource

Assemble a reviewable release PR — versions computed from accumulated .changes/ intents, changelogs written, narrative synthesized, release meta emitted. Use when the user says "cut a release", "release the core train", "assemble a release", or invokes /release. Releases are deliberate; never run th

Install

mkdir -p .claude/skills/release-mattstyles && curl -L -o skill.zip "https://agentskills.codes/api/skills/download/16758" && unzip -o skill.zip -d .claude/skills/release-mattstyles && rm skill.zip

Installs to .claude/skills/release-mattstyles

Activation

This is the description your AI agent reads to decide when to run this skill — the better it matches your request, the more reliably it fires.

Assemble a reviewable release PR — versions computed from accumulated .changes/ intents, changelogs written, narrative synthesized, release meta emitted. Use when the user says "cut a release", "release the core train", "assemble a release", or invokes /release. Releases are deliberate; never run this automatically on merge.
326 chars✓ has a “when” triggerlonger than Claude Code's old 250-char listing cap (fine on current versions)

About this skill

Release assembly

Deterministic mechanics are scripts (internal/release, fixture-tested); judgment — the release narrative — is yours. The release PR is the review surface and merge is the commitment point ([[0004-release-strategy]]).

Flow

  1. Preflight — on a clean main, run:

    bun internal/release/src/cli.ts status

    If no train has pending intents, stop and say so. Confirm with the user which train(s) they expect to depart if the status disagrees with them.

  2. Branch — release/<train>-<next-version> (both trains departing → release/<date-slug>).

  3. Assemble — run:

    bun internal/release/src/cli.ts assemble

    This applies versions to every train member, prepends per-package CHANGELOG.md sections from intent prose, consumes the intent files, and emits releases/<train>-<version>.json plus a narrative skeleton releases/<train>-<version>.md.

  4. Narrate — replace the skeleton body with a release narrative that will become the GitHub Release body. Synthesize from three sources; do not concatenate entries:

    • the intent prose (consumer-addressed, already reviewed),
    • manifest diffs (git diff main -- '**/urban-manifest.json') for the factual API surface changes,
    • linked PR descriptions (gh pr list --search <sha> / gh pr view) for motivation, migration detail, and screenshots — best effort only: if offline, the narrative from intents + manifests alone is still correct.

  5. Verify — mise run manifest-check and hk check --all must pass; changelog and version diffs should match the status output from step 1.

  6. PR — commit everything (versions, changelogs, consumed intents, releases/ meta + narrative) and open a PR titled Release: <train> v<version> describing what departs and why now. Merging the release PR is what triggers the publish plane — do not merge without the user's say-so.

Search skills

Search the agent skills registry