Templates
Copy-paste artifacts for agent-driven work. Each template is opinionated and fill-in-able: placeholders in <angle brackets>, no essays. Prefer these over inventing structure mid-session — agents follow a filled template more reliably than a vague prompt.
By category
| Category | Page | Reach for when… |
|---|---|---|
| Repo docs | Repo Docs | Standing up or hardening agent memory files (CLAUDE.md, ARCHITECTURE.md, etc.) |
| Specs & planning | Specs & Planning | Turning intent into something an agent can implement without inventing requirements |
| Checklists | Checklists | Human gates: review, deploy, security, readiness |
| Agent operations | Agent Operations | Handing off sessions, resetting context, splitting multi-agent work |
| Process | Process | ADRs, releases, postmortems, architecture reviews |
All templates
| Template | Category | When to use |
|---|---|---|
| CLAUDE.md | Repo docs | Every Claude Code repo — standing orders under ~60 lines |
| AGENTS.md | Repo docs | Multi-tool repos; source of truth when Claude Code is not the only agent |
| README.md | Repo docs | Human onboarding; keep agent-critical facts out of here |
| ARCHITECTURE.md | Repo docs | System shape agents must not invent around |
| TASKS.md | Repo docs | Living work queue the agent updates in-PR |
| DECISIONS.md | Repo docs | Lightweight decision log when full ADRs are overkill |
| CODING_STANDARDS.md | Repo docs | Non-lintable conventions agents keep violating |
| TESTING.md | Repo docs | How to run tests, what "done" means for verification |
| DEPLOYMENT.md | Repo docs | Deploy path, envs, rollback — agents invent this otherwise |
| SECURITY.md | Repo docs | Threat model pointers and never-dos for secrets/auth |
| PROJECT_CONTEXT.md | Repo docs | Product/domain facts that don't belong in CLAUDE.md |
| New project brief | Specs | Before scaffolding; forces scope and non-goals |
| MVP spec | Specs | First shippable slice with acceptance criteria |
| Feature spec | Specs | Any non-trivial feature into an existing codebase |
| Bug report | Specs | Repro + expected vs actual before asking an agent to "fix" |
| Refactor plan | Specs | Behavior-preserving change with blast radius named |
| Task breakdown | Specs | Splitting a spec into agent-sized units |
| Sprint plan | Specs | Team cadence with agent capacity explicit |
| Code review checklist | Checklists | Human review of any PR (AI or not) |
| AI-generated PR checklist | Checklists | Extra gates for agent-authored diffs |
| Deployment checklist | Checklists | Pre-prod and go-live |
| Security review checklist | Checklists | Auth, secrets, tenancy, injection surfaces |
| Testing plan | Checklists | What to verify before merge |
| Production readiness review | Checklists | First production cut or major subsystem |
| Agent handoff note | Agent ops | Ending a session so the next one continues cleanly |
| Context reset note | Agent ops | Mid-task thrash — compact and restart |
| Repository onboarding note | Agent ops | First session in an unfamiliar repo |
| Multi-agent task split | Agent ops | Parallel agents with non-overlapping ownership |
| ADR | Process | Cross-cutting design choice that must survive sessions |
| Release notes | Process | Shipping a version humans and agents can cite |
| Postmortem | Process | Incident or agent-caused outage write-up |
| Architecture review notes | Process | Structured review of a proposed or existing design |
32 templates. Prefer linking the filled artifact from CLAUDE.md over pasting it into every prompt.
Related
- Docs as Agent Memory — why these files exist
- CLAUDE.md & AGENTS.md — how to keep instruction files sharp
- Writing Effective Specs — what belongs in a spec vs a template
- Decision Records — ADR practice beyond the template
- Prompt Library — prompts that consume these templates