setup-playbook
Skill0 starsChanged 3 months agoMITView on GitHub
Bootstrap the current repository with the playbook's infrastructure — CI gates, agent review workflow, CLAUDE.md, CONTEXT.md, ADR directory, git hooks. Use the day a project stops being throwaway ("I'd be annoyed if this broke").
- Reads credentials
In this repository12 other files it gives its agents.
- templates/CLAUDE.mdCLAUDE.md
- build-supervisedSkill
- decomposeSkill
- diagnoseSkill
- handoffSkill
- interrogateSkill
- prdSkill
- respond-to-reviewSkill
- specSkill
- spikeSkill
- tddSkill
- verify-liveSkill
The file
We split this skill into the rules it gives an agent
2 rules. About 0 are common, 1 is distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 33
keep the tool allowlist read-only: the reviewer comments, it never pushes
- Very specific: line 62
Verify the installation: run the gates locally under the repo's pinned runtime — read .node-version / .tool-versions / equivalent and …
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Setup Playbook
Installs the standing infrastructure the three loops depend on. Run once per repository. Idempotent: never overwrite an existing file — show a diff and merge instead.
Steps
- Inspect the repo first. Detect: package manager, language, formatter/ linter, test runner, existing CI, existing CLAUDE.md/CONTEXT.md. Every template below gets adapted to what's found, not pasted blind.
- CI gates — from
templates/github/workflows/ci.yml:- format + lint check, typecheck, and dependency-free unit tests
- fill in the repo's real commands; delete the placeholder comments
- gates must be FAST (minutes) — anything slow runs elsewhere
- Expect the format/lint gate to FAIL on the existing default branch if the repo has only ever had pre-commit hooks: staged-files-only hooks let unformatted content arrive via merges and web edits. Run the gates against the default branch first and clear the drift in the install PR itself, as its own commit — the gate must start green.
- Agent review workflow — from
templates/github/workflows/agent-review.yml:- reviews every PR from a fresh context; always posts a summary
- tell the user to add the
ANTHROPIC_API_KEYsecret (repo settings) — the workflow is inert without it, and you cannot add secrets for them - keep the tool allowlist read-only: the reviewer comments, it never pushes
- Git hooks — from
templates/lefthook.yml(or the repo's existing hook manager): pre-commit format-fix + re-stage of staged files only. CLAUDE.md— fromtemplates/CLAUDE.md. Fill in: launch commands (ask the user or read compose files/scripts), test commands, and any framework guard blocks that apply. If one exists, merge sections rather than replacing.CONTEXT.md— fromtemplates/CONTEXT.md. Seed the glossary with terms already in the codebase (status enums, job names, user-facing nouns) — an empty glossary teaches nothing. Create the shared-resource registry and pre-fill current claims (existing migration numbers, ports, queue names).- ADR directory —
docs/adr/with the template and ADR-0001 (fromtemplates/docs/adr/). PRINCIPLES.md— fromtemplates/PRINCIPLES.md. If the operator is a non-coding owner, fill it WITH them now (product principles, quality bar, ask-first list, risk posture) — it is the file that makes their judgment enforceable by agents and reviewers alike. If standing rules already live in-repo (a rules section in CLAUDE.md, a sacred-files list), do NOT create a duplicate: point the review workflow's prompt at the existing sections by name and record them as this repo's principles file. One source of truth beats the template.- Verify the installation: run the gates locally under the repo's
pinned runtime — read
.node-version/.tool-versions/ equivalent and activate it explicitly (version managers often don't hook non-interactive shells, and a stale global runtime fails in ways that look like repo bugs). Then open a trivial PR if the user agrees, and confirm the review workflow fires.
Report
End with: what was installed, what was skipped (and why), what needs the user's hands (secrets, branch protection), and the one-line reminder of the per-feature loop: interrogate → spec → decompose → generate → verify-live → respond-to-review.
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
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.

