Чек-лист релиза
Проводит проверки до и после релиза: миграции, конфиг и env, флаги, обратная совместимость, мониторинг, план отката, release notes. Используйте перед выкаткой или тегом релиза.
release-checklistУстановка
Выполните одну команду — она сохранит SKILL.md в папку скиллов агента. Перезапустите агента, и он подхватит скилл сам.
mkdir -p ~/.claude/skills/release-checklist && curl -fsSL https://anymodel.org/en/skills/devops/release-checklist/SKILL.md -o ~/.claude/skills/release-checklist/SKILL.md
mkdir -p ~/.agents/skills/release-checklist && curl -fsSL https://anymodel.org/en/skills/devops/release-checklist/SKILL.md -o ~/.agents/skills/release-checklist/SKILL.md
New-Item -ItemType Directory -Force "$HOME\.claude\skills\release-checklist" | Out-Null; Invoke-WebRequest https://anymodel.org/en/skills/devops/release-checklist/SKILL.md -UseBasicParsing -OutFile "$HOME\.claude\skills\release-checklist\SKILL.md"
New-Item -ItemType Directory -Force "$HOME\.agents\skills\release-checklist" | Out-Null; Invoke-WebRequest https://anymodel.org/en/skills/devops/release-checklist/SKILL.md -UseBasicParsing -OutFile "$HOME\.agents\skills\release-checklist\SKILL.md"
Другой агент: сохраните файл в его папку скиллов или дайте ему ссылку на файл: https://anymodel.org/en/skills/devops/release-checklist/SKILL.md
SKILL.md
Release Checklist
Assess whether a set of changes is safe to ship and what to verify afterwards. You produce a go/no-go report; the user decides and performs the deploy. Never deploy, tag, publish or push without explicit confirmation.
Step 1: Define the release scope
git fetch origin
git log --oneline <last-release-tag>..HEAD
git diff --stat <last-release-tag>..HEAD
List the changes grouped by area. Identify the risky ones: schema changes, auth, payments, background jobs, public API, dependency upgrades, infrastructure files.
Step 2: Pre-release checks
Code and tests
- CI is green on the exact commit to be released. Run the test suite locally if CI is unavailable.
- No leftover debug code,
TODO: remove, or temporary flags forced on. - Dependency changes reviewed (lock file diff, major bumps, new licenses).
Database migrations
- List every new migration. For each: is it reversible, does it lock big tables, does it need a backfill?
- Migrations must be compatible with the previous app version, because old and new code run side by side during a rollout. Destructive steps (drop column/table) belong in a later release.
- Decide the order: migrate before deploy (additive) or after (cleanup).
Config and environment
- Diff config and env usage: new variables read in code, renamed or removed ones.
git diff <last-release-tag>..HEAD -- .env.example docker-compose* '*.yaml' '*.yml'
git grep -nE "process\.env\.|os\.environ|getenv" -- <changed-paths>
- Every new variable exists in the target environment (staging and production) with a correct value. Do not print secret values; report only names and whether they are set when you can check.
- Secrets that must be rotated are called out.
Feature flags
- New risky behavior is behind a flag, default off, with a named owner and a removal plan.
- Flag states in production match the plan for this release.
Backward compatibility
- Public API, webhooks, queues and file formats: no breaking change without versioning or a deprecation path.
- Clients on older versions (mobile apps, cached frontends) still work.
- Message and event schemas are readable by both old and new consumers.
Monitoring and alerts
- Key dashboards identified: error rate, latency, saturation, business metric (orders, signups).
- Alerts exist for the new code paths; logs include enough context to debug.
- Someone is available to watch the release and has the access to roll back.
Rollback plan
Write it down before shipping:
- Exact rollback command or procedure and who runs it.
- Whether migrations can be undone; if not, what the fallback is (backup taken, restore time).
- The signal and threshold that trigger a rollback (for example 5xx above 2% for 5 minutes).
Release notes
- User-facing summary, upgrade steps, known issues, and breaking changes, derived from the commit list rather than invented.
- Version number follows the project's scheme (semver if used).
Step 3: Report go/no-go
Release: <version> commit: <sha> range: <from>..<to>
Verdict: GO | GO with conditions | NO-GO
Blockers: ...
Conditions: ...
Risky changes: ...
Migrations: ... (order, reversibility)
Config to set before deploy: names only
Rollback plan: ...
Draft release notes: ...
Step 4: Post-release verification
After the user confirms the deploy:
- Check the version actually running (health or version endpoint, image tag).
- Watch error rate and latency for 15-30 minutes against the pre-release baseline.
- Run a smoke test of the critical user paths (login, checkout, main API call).
- Confirm migrations applied and background workers are healthy.
- Report the result and update release notes or the changelog; remind the user to remove stale flags later.
Pitfalls
- Do not release on a Friday evening or before a holiday without telling the user the risk.
- Do not mix unrelated large changes into one release if they can ship separately.
- A green CI does not prove production config is correct; check it explicitly.
- Do not run production migrations yourself; give the command and wait for confirmation.
Source: AnyModel skills catalog · License: CC BY 4.0
Подключить через AnyModel API
Скилл работает на той модели, которую использует агент. Подключите Claude Code, Codex или Cursor к AnyModel и запускайте его на Claude, GPT, Gemini и других моделях по одному ключу — оплата за токены, без подписки.
AnyModel