本文へ移動
cccskills
無料GitHub で公開

batch-artifact-verification

Use when a batch (desktop or mobile) release reaches awaiting_verdict or was rolled back - read-only checks that the published artifact really exists and is served, and the exact yank step for a bad tag

インストール方法を見る

含まれるファイル(1)

  • SKILL.md4.1 KB

SKILL.md(原文)

インストールする前に、エージェントに与えられる指示の中身を確認できます。

Batch artifact verification

A batch release (desktop, mobile and other human-cut components) has no bound runtime environment — see post-deploy-verification's last section. Its evidence is whether the thing it built actually got published and is what users will fetch next, per executor.

github_actions / local

run_terminal("gh release view <tag> --json tagName,isDraft,isPrerelease,isImmutable,publishedAt,assets")

Expect isDraft: false, the platform's assets present with size > 0, and the update-feed file if the app ships one (e.g. latest-mac.yml). If gh is unavailable but the repository is public, fall back to fetch_url https://api.github.com/repos/<owner>/<repo>/releases/tags/<tag>.

If the repository brief names an update feed or a Homebrew cask URL, fetch_url it and confirm it names the new version — that is what actually reaches an existing install, not the GitHub Release page by itself.

For a local executor, the exit code is the first signal: local_run.exit_code == 0, with local_run.tail showing the publish step actually ran (not just that the build compiled).

store

Every entry in store_builds[] must have build != baseline_build and carry no error. A baseline_build of "?" means the baseline was never confirmed — say so rather than treating it as a pass. "Deployed" here means the build reached the internal/testing channel; promoting it to production is a human step taken in the store console, and your finish note says so explicitly rather than implying the rollout is live.

Yanking a bad batch release

This is the first manual_steps item on a batch rollback (rollback-runbook), and you perform it yourself — it is the one write release-terminal-scope allows:

run_terminal('gh release edit <tag> --prerelease --title "<name> [YANKED]"')

This works on an immutable release (title, notes and prerelease/latest stay editable; assets and the tag itself do not) and removes it from GitHub's "latest" resolution (the most recent non-prerelease, non-draft release), which most update feeds follow via /releases/latest. Then point "latest" back at the previous good release:

run_terminal("gh release edit <previous-good-tag> --latest")

Record the exact commands you ran and their output as evidence on the card — this is not a suggestion for a human to carry out, it is work you did and are reporting.

Never delete the release or its tag, and never git push --delete it — an immutable tag cannot be reused (GitHub Docs: immutable releases), and SemVer §3 holds here too: a released version's contents are never modified, only marked bad. A cask/update-feed PR pointing at the new tag, or halting a store rollout, are human steps outside gh release's reach — report them, naming exactly what needs to change and where.

✅ "Yanked v2.4.1: gh release edit v2.4.1 --prerelease --title \"TaskTrooper 2.4.1 [YANKED]\" (exit 0); restored latest to v2.4.0: gh release edit v2.4.0 --latest (exit 0). gh release view v2.4.1 now shows isPrerelease: true. Left for a human: the Homebrew cask formula still points at v2.4.1 — needs a PR bumping it back to v2.4.0."

❌ "Release looks bad, someone should pull it." — no command run, no evidence, and it was this agent's job to run it, not a human's.

Common Mistakes

  • Treating a store build reaching the internal channel as "released to users" — it is not, until a human promotes it.
  • Deleting a release or tag instead of marking it [YANKED] and moving --latest.
  • Reporting a yank as done without the command output that proves it happened.
  • Checking only isDraft/isPrerelease and skipping the update feed — a feed that still points at the bad tag means existing installs keep fetching it even after the GitHub Release itself is marked yanked.

レビュー

まだレビューはありません。使ってみた感想をお寄せください。

同じリポジトリのスキル

概要と使いどころ

Use when writing acceptance criteria for a task - express each as an observable Given/When/Then that QA can execute, including negative cases

日本語の概要は準備中です。原文の説明を表示しています。

makifbaysal/tasktrooper1122026年10月10日 更新

Use when the diff adds or changes an endpoint, resolver, RPC, job or query that takes an object id, a role check, a request binding or a tenant filter - BOLA/IDOR, function-level authorization, mass assignment and tenant scoping

日本語の概要は準備中です。原文の説明を表示しています。

makifbaysal/tasktrooper1122026年10月10日 更新

Use on every UI change - semantic HTML, labels for controls, keyboard-navigable dialogs/menus, visible focus, and never color as the only signal

日本語の概要は準備中です。原文の説明を表示しています。

makifbaysal/tasktrooper1122026年10月10日 更新

Use when a task changes any screen, form, dialog, menu or control - Lighthouse/axe scan of the changed screens, a keyboard walk, and the thresholds that fail a task

日本語の概要は準備中です。原文の説明を表示しています。

makifbaysal/tasktrooper1122026年10月10日 更新

How to work a task returned with review, QA or UAT findings. Use when a task is in need_revision or PR review comments are in your context.

日本語の概要は準備中です。原文の説明を表示しています。

makifbaysal/tasktrooper1122026年10月10日 更新

Use when deciding whether a request needs an analiz task before implementation - the conditions that require the architect's analysis versus going straight to implementation

日本語の概要は準備中です。原文の説明を表示しています。

makifbaysal/tasktrooper1122026年10月10日 更新

makifbaysal のスキルをすべて見る

このスキルの問題を報告する