メニュー
Release My Assistant — run gates, bump version, update changelog, commit, tag, and push. Use when asked to "release", "ship", "bump and push", or "cut a release".
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
| name | release |
| description | Release My Assistant — run gates, bump version, update changelog, commit, tag, and push. Use when asked to "release", "ship", "bump and push", or "cut a release". |
End-to-end release workflow for My Assistant.
Run all gates in parallel:
python3 scripts/validate.py
python3 scripts/validate_fixtures.py
Requires pip install -r requirements.txt if PyYAML is not already installed.
Stop if any gate fails. Fix issues before proceeding.
For minor or major releases, also run the connector smoke subset (docs/guide/connector-smoke-tests.md) and inbox smoke subset (evals/README.md#smoke-subset) before cutting the release.
Read the current version from .claude-plugin/plugin.json. Ask the user which semver component to bump if not specified:
| Bump | When |
|---|---|
| patch | Bug fixes, test/fixture updates, docs |
| minor | New skills, commands, agents, or user-visible features |
| major | Breaking changes to commands, skills, profile schema, or hooks |
Default to patch if the user says "release" without specifying.
Update the version field in both manifest files — keep them in lockstep per CONTRIBUTING.md:
.claude-plugin/plugin.json.cursor-plugin/plugin.jsonThese are the only version sources of truth. There is no package.json version to sync.
In CHANGELOG.md:
[Unreleased] accurately lists user-visible changes for this release.[Unreleased] to [X.Y.Z] - YYYY-MM-DD (today's date).[Unreleased] section at the top with ### Added, ### Changed, and ### Fixed subsections per existing style.git add -A
git commit -m "chore(release): bump plugin to <new-version>"
git tag v<new-version>
git push && git push --tags
Commit message style: match existing convention — chore(release): bump plugin to X.Y.Z or chore(release): vX.Y.Z — <short summary> (see git log for examples).
Only commit when the user explicitly asks. Privacy check: no real profiles, .env, or *.local.md in the commit.
If a hook fails, fix the issue and create a new commit (never amend).
.claude-plugin/plugin.json and .cursor-plugin/plugin.json — the version field in each (must match). There is no build step; this is a file-based plugin.
python3 scripts/validate.py passespython3 scripts/validate_fixtures.py passesplugin.json manifests bumped to the same versionCHANGELOG.md updatedv<version> createdOnboarding for the assistant plugin. Activate when the user says "/assistant:setup", "set me up", "configure my assistant", "help me get started", or when no profile exists per rules/paths.md. Writes profile, policies, memory (AGENTS.md hot cache + memory/ tree), and install config.
Propose meeting times, check conflicts, protect calendar time with buffer/prep/follow-up/habit blocks, and draft invites and scheduling replies. Activate when the user asks to schedule something, find a time, check for conflicts, protect their calendar, or draft an invite. Proposes and drafts only — never books, moves, or declines without approval.
A short, scannable morning briefing — today's calendar, priority unread, follow-ups going cold, and what needs attention. Activate when the user says "/assistant:brief", "morning briefing", "what's my day look like", or on a morning schedule.
Produce a pre-meeting brief — who's in the room, what they do, last contact, and what to prepare. Activate when the user says "/assistant:meeting prep", "prep me for my next meeting", "who am I meeting", or on a meeting-prep schedule.
Draft email, chat, and letter replies in the user's own voice. Activate when triaging needs-reply mail, when the user says "/assistant:draft email|chat|letter", "/assistant:email draft", "draft a reply", or pastes a message they need to answer. Creates drafts only — never sends.
Capture feedback on email reply drafts and propose profile voice updates. Activate when the user says "/assistant:email feedback", rates a draft as good, light edit, or heavy rewrite, or pastes what they actually sent after editing a draft. Proposes profile diffs for approval — never writes profile.md directly.