agentleFS
0xfurai0xfurai/peekaping/.agents/skills/release/SKILL.md

release

Skill1.2k starsChanged 15 days agoMITView on GitHub

Cut a Peekaping release: check main and the changelog, dispatch the Release workflow, watch it, verify the tag, release notes and images, then open the follow-up PR. Also builds release candidates (X.Y.Z-rc.N).

196 other projects use a version of this.

Trending

More talk lately.thin dataabout the whole repository

Dec 2025: 0 mentionsJan 2026: 0 mentionsFeb 2026: 0 mentionsMar 2026: 0 mentionsApr 2026: 0 mentionsMay 2026: 0 mentionsJun 2026: 0 mentionsJul 2026: 0 mentionsAug 2026: 0 mentionsSep 2026: 0 mentionsOct 2026: 1 mentionNov 2026: 0 mentions
Dec 2025Jun 2026Nov 2026
Mentions a month of the 0xfurai/peekaping repository that our weekly crawl found: only 2 in all, so this is a hint, not a trend. There is no adoption data for this file yet.

Most recently taken up by

Used with

Projects with it also use create-agent-adapter, changelog-style and 10 more.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • MicrosoftSkill

    create-agent-adapter

    Technical guide for creating a new Paperclip agent adapter. Use when building a new adapter package, adding support for a new AI coding tool (e.g. a new CLI agent, API-based agent, or custom process), or when modifying the adapter system. Covers the required interfaces, module structure, registration points, and conventions derived from the existing claude-local and codex-local adapters.

    Also in 5 projects that use this

  • GrinvSkill

    changelog-style

    Turn a commit range into CHANGELOG.md entries following common-changelog style. Use when finishing a PR or cutting a release, to write/fix the [Unreleased] section's entries — moving [Unreleased] under a new dated version heading is a separate, automated step (see the release skill / scripts/sync-version.ts), not this skill's job.

    Also in 4 projects that use this

  • GrinvSkill

    docs-consistency-check

    Check README/manifest.json/server.json/CHANGELOG.md/AGENTS.md and docs/*.md for drift against the actual registered tools and source. Use after adding, renaming, or removing a tool, or as part of a live-audit pass.

    Also in 4 projects that use this

  • Microsoft AzureSkill

    component-release

    Prepares a verifiable Agent Landing Zone UI release. Use for release branches, VERSION, changelog entries, tags, release notes, and compatibility evidence; never publish without explicit human approval.

    Also in 4 projects that use this

  • Microsoft AzureSkill

    documentation-consistency

    Keeps AI Landing Zone contributor, user, and operator documentation aligned with shipped Bicep behavior. Use for parameters, defaults, outputs, topology, deployment flow, operations, or releases.

    Also in 8 projects that use this

  • rivet-devSkill

    launch-graphics

    Generate agentOS marketing/launch graphics (launch & social heroes, code-snippet cards) from the committed generators in docs-internal/graphics. Use when asked to make, update, or render an agentOS launch image, social image, "agentOS Apps" graphic, or technical code-snippet image.

    Also in 3 projects that use this

  • +3 more from Grinv

The file

We split this skill into the rules it gives an agent

6 rules. 0 are common, 3 are distinctive, and 3 are very specific to this project.

  • 0

    Common

    Most of the 186 edited versions we read keep these.

  • 3

    Distinctive

    This author's own choices. Worth a look.

  • 3

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Never pass the workflow's changelog input.

    line 24
  2. Distinctive:

    Every migration listed should have an entry that says what it does to an existing database.

    line 45
  3. Very specific:

    Only when step 2 found gaps.

    line 55
  4. Very specific:

    Keep only the ### headings.

    line 59
  5. Distinctive:

    Don't report the release out until each of these is checked:

    line 87
  6. Distinctive:

    Ask which branch to build: main, or the branch of a PR the user wants to try.

    line 116

“Common” is measured: kept by at least 6 in 10 of the 186 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Release

The Release workflow (.github/workflows/docker-release.yml) is started by hand with a version. In 15 to 40 minutes it:

  1. builds every image for amd64 and arm64 and pushes them to GHCR and Docker Hub;
  2. merges the multi-arch manifests (and latest) and runs the e2e suite against the pushed images;
  3. bumps version.go, package.json and version.ts, moves [Unreleased] in CHANGELOG.md under the new version, and pushes that commit straight to main;
  4. creates the GitHub Release, with its tag on that commit;
  5. pushes the Helm chart to oci://ghcr.io/0xfurai/charts/peekaping, with version and appVersion set to the release, so it installs that release's images;
  6. deploys the demo.

It runs no unit tests, and none of it can be taken back cleanly: the images and the release are public and the commit is on main. So the work is in the steps before the dispatch.

Rules

  • The user decides. You prepare, check, and dispatch only after their explicit go-ahead on the exact version.
  • Never pass the workflow's changelog input. It is pasted into a shell script, and the entries are full of backticks, which the shell would run. Left empty, the workflow reads [Unreleased] itself.
  • From the dispatch until the GitHub Release exists, nothing merges into main. The bump job pushes to main after the e2e suite, from the commit the run started on, and fails if main moved since; a merge after that lands in the tag, because the release job cuts the tag at the head of main.
  • Tags have no v prefix: 0.0.47.
  • Other sessions share the main checkout. Files change in a worktree under .claude/worktrees/; git fetch, never switch branches in the main checkout.

1. Choose the version

Take it from the argument. With none, propose the next patch after the last tag (git tag -l | sort -V): every release so far is 0.0.N. An X.Y.Z-rc.N version goes to release candidates.

2. Preflight

From the repo root (or any worktree of it):

.agents/skills/release/scripts/preflight.sh <X.Y.Z>

It reads origin/main and reports: the version is new, CI on main, the [Unreleased] entries, PRs merged since the last tag that no entry mentions, migrations added since, and whether the Helm chart's bundled PostgreSQL and Redis are behind the latest. BLOCK lines exit 1.

  • A red or still-running CI is a blocker: the release workflow won't catch what Test and Lint and Playwright Tests catch.
  • The PR list is a worklist, not a verdict. Entries cite PR numbers, issue numbers or nothing, so read the entries for each PR on it. Tests, CI and refactors with no visible change need no entry.
  • Every migration listed should have an entry that says what it does to an existing database.
  • A WARN Helm chart: line never blocks. It says one of three things:
    • ... are behind the latest: Bitnami has published newer PostgreSQL or Redis subcharts or images than the chart pins. Run scripts/update-chart-deps.sh --write in a PR, before the release or after it.
    • ... has a new major version: a subchart or an image has another major version than the chart pins. --write refuses it and writes nothing, not even for the lines that are only behind, because Bitnami's major versions are breaking. It is a ticket of its own: tell the user, and don't run --write.
    • couldn't check ..., or origin/main has no scripts/update-chart-deps.sh: nothing is known to be behind. Tools (helm, curl, jq), files on origin/main or access to Docker Hub are missing. Fix that and run the preflight again, or run scripts/update-chart-deps.sh by hand to see its report. --write is not the next step.

Done when: no blockers, and every PR on the list has an entry or a reason not to.

3. Curate the changelog

Only when step 2 found gaps. In a worktree (git worktree add .claude/worktrees/release-<X.Y.Z> -b chore/release-<X.Y.Z>-changelog origin/main), edit [Unreleased] in CHANGELOG.md:

  • One entry per user-visible change, under Added, Changed or Fixed, in the style of its neighbours: type: what the user sees now (#issue, #pr) (thanks @author). Read the PR (gh pr view <n>) for what changed; credit authors the way the neighbouring entries do.
  • A change that needs the user to act on upgrade, such as a migration, a new default or a moved setting, says so in the entry, starting Upgrade note:.
  • Keep only the ### headings. The workflow copies [Unreleased] with awk, which drops every other line that starts with #.

Open one PR as in STACK.md, without a ticket key: docs: complete the changelog for <X.Y.Z>. The user merges it. Then run the preflight again.

4. Get the go-ahead

Ask with the brief in PLAN.md, one AskUserQuestion: dispatch <X.Y.Z> or wait. The brief gives the version and the last tag, the origin/main commit, what [Unreleased] says in a few plain lines, the upgrade notes and migrations, and anything the preflight flagged and how it settled. Ask the user not to merge into main until you report the release is out.

Done when: the user said to dispatch that version. If origin/main moved since you asked, run the preflight again and re-confirm.

5. Dispatch

gh workflow run docker-release.yml --ref main -f version=<X.Y.Z>
gh run list --workflow=docker-release.yml --limit 1 --json databaseId,headSha,status

The run takes a few seconds to appear. Check that its headSha is the commit the user approved. Then watch with gh run watch <id> --exit-status as a Bash run_in_background; its exit wakes you. The jobs run in this order: build-images and build-single-image (both per architecture), create-manifest, playwright-tests, bump-version, release, then publish-chart and deploy-demo side by side.

6. If the run fails

Find the cause: gh run view <id> --log-failed. Then see whether the bump landed: git fetch origin && git log origin/main --author=github-actions -3 --format='%h %s', looking for chore: bump version to <X.Y.Z> and update changelog.

  • No bump commit. main is untouched, but once create-manifest has passed the version tags and latest are public. A flake or an outage gets one gh run rerun <id> --failed, which reruns the failed jobs and the ones after them on the same commit. A real failure stops here: tell the user the cause. The fix is a normal PR; then preflight and dispatch the same version again, which overwrites the manifests (nothing rolled [Unreleased]). If bump-version failed because main moved, a rerun fails the same way, since the job starts from the commit the run started on; dispatch again from the new head, after the go-ahead of step 4.
  • Bump commit on main. Only release, publish-chart and deploy-demo run after it. Never dispatch the same version again: the changelog step would roll an empty [Unreleased] into a second ## [<X.Y.Z>] heading. A flake gets one gh run rerun <id> --failed. A real failure goes to the user with the brief in PLAN.md: the images and the rolled changelog exist, the release doesn't, and the way out is to fix forward and release the next patch.

7. Verify

Don't report the release out until each of these is checked:

  • Run. gh run view <id> shows success. deploy-demo doesn't count: its deploy step is continue-on-error, so a failure shows only as a "Demo not updated" warning on the run. Read that job's log: it prints the demo's image: lines, which should end in <X.Y.Z>.
  • Main. git log origin/main --author=github-actions -1 is the bump commit. git show origin/main:CHANGELOG.md | head -30 shows an empty [Unreleased] and ## [<X.Y.Z>] - <today> under it.
  • Tag and release. gh release view <X.Y.Z> --json tagName,isPrerelease,url,body: not a pre-release, the body starts with the notes you reviewed. git rev-parse <X.Y.Z>^{commit} is the bump commit; if not, say which commits landed after it.
  • Images. For each of web api migrate probe, docker buildx imagetools inspect 0xfurai/peekaping-<name>:<X.Y.Z> lists linux/amd64 and linux/arm64. The single-mode image has no suffix: the same check on 0xfurai/peekaping:<X.Y.Z>. (peekaping-bundle-postgres gets no new tags: its last is 0.0.47.) For one of them, --format '{{.Manifest.Digest}}' is the same for :<X.Y.Z> and :latest. Through 0.1.0, the old name 0xfurai/peekaping-bundle-sqlite gets the same tags, so also check that its --format '{{.Manifest.Digest}}' equals the one of 0xfurai/peekaping at both <X.Y.Z> and latest. After 0.1.0 the old name gets no tags, so skip this (PP-321). This checks Docker Hub; GHCR pushes come from the same green create-manifest job, and an anonymous inspect of GHCR can answer 403.
  • Helm chart. helm show chart oci://ghcr.io/0xfurai/charts/peekaping --version <X.Y.Z> shows version and appVersion of <X.Y.Z>. After the first release that publishes it, the GHCR package charts/peekaping is private: in its package settings on GitHub, link it to the repository and make it public, or the command answers 401 for everyone else.
  • First release with the new image name (0.0.48), one-time steps. These are manual, because a new GHCR package starts private and the workflow can't change that:
    • After the first push of peekaping (an RC or the release, whichever comes first): in the settings of the GHCR package peekaping on GitHub, link it to the repository and make it public. Until then an anonymous docker pull ghcr.io/0xfurai/peekaping:<version> (the RC or release version: an RC pushes no latest) answers 401 or 403. The e2e job logs in, so the release itself doesn't need it. Check on Docker Hub that the repository 0xfurai/peekaping is public and has a description.
    • After the release: on the demo host, change the image: line in the compose file to 0xfurai/peekaping:<X.Y.Z>. deploy-demo.yml only seds the tag, so it keeps the old name until you do. Update the RepoCloud template to the new image name. Point the Docker Hub and GHCR descriptions of peekaping-bundle-sqlite at peekaping.
    • The old name peekaping-bundle-sqlite gets tags through 0.1.0. After it, the follow-up ticket removes its lines from docker-release.yml (each carries a comment with PP-321) and this check.

8. Open the OpenAPI PR

A known gap in the workflow: the bump changes the API's version, which is written into apps/server/docs/openapi.json and openapi.yaml, but the workflow doesn't regenerate them, and its push starts no CI. The first PR that touches apps/server or apps/web then fails bun run api:check.

In a worktree off the updated origin/main, run bun run api:generate. The diff should be the version line in those two files and nothing else. If go doesn't run on this machine, change that one line in each file by hand; CI's api:check confirms it. Commit chore(server): regenerate the OpenAPI docs for <X.Y.Z>, open the PR as in STACK.md without a ticket key, and give the user the link: it should merge before any other PR. Delete this step when the workflow regenerates the docs itself.

9. Wrap up

Tell the user: the release URL, the run URL, what you verified, anything to look at (the demo, an upgrade note), and the OpenAPI PR. Remove the release worktrees once their PRs merge: git worktree remove .claude/worktrees/release-<X.Y.Z>.

Done when: the GitHub Release is out with the right notes, both architectures are on Docker Hub, main has the bump, and the OpenAPI PR is open.

Release candidates

X.Y.Z-rc.N runs release-candidate.yml. It runs the web lint and build and the server tests first, pushes every image tagged with the RC version only (never latest), the single-mode image as peekaping and not under the old name, and publishes a pre-release at the head of the branch it ran on. It leaves CHANGELOG.md and the version files alone, so the preflight and changelog steps (2, 3), the bump case of step 6 and step 8 don't apply.

  1. Next number: git tag -l 'X.Y.Z-rc.*' | sort -V | tail -1, plus one; rc.1 for the first.
  2. Ask which branch to build: main, or the branch of a PR the user wants to try. The RC is public, so get the go-ahead on the version and branch, as in step 4.
  3. gh workflow run release-candidate.yml --ref <branch> -f version=<X.Y.Z-rc.N>, with no changelog input, then watch as in step 5.
  4. Verify as in step 7: the run, gh release view shows a pre-release, and the images carry both architectures. On a PR branch the workflow comments the images on the PR. The RC that first pushes peekaping creates its GHCR package: do the one-time steps for the new image name from step 7 now, so the package is public before the release.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.