operating-doctrine

---
name: operating-doctrine
description: Apply the Agent Kit operating doctrine in Claude Code. Use when the task needs safety, planning, verification, or cross-provider consistency.
when_to_use: Use for complex tasks, architecture changes, provider-layer updates, or any task where process discipline matters.
argument-hint: "[task or scope]"
---

# Operating Doctrine

Apply Agent Kit's operating loop in Claude Code.

## Required Flow

```text
EXPLORE -> PLAN -> EXECUTE -> VERIFY -> REPORT
```

## Rules

1. Gather read-only evidence before changing state.
2. Define goal, scope, blast radius, permissions, verification, and rollback.
3. Execute only approved or policy-allowed changes.
4. Serialize mutations on shared state.
5. Verify with fresh evidence before claiming completion.
6. Report changed files, verification, risks, and remaining work.

## Canonical References

- Master doctrine: `AGENTS.md`
- Claude Code entrypoint: `CLAUDE.md`
- Claude Code layer map: `.claude/ARCHITECTURE.md`
- Skill inventory: `skills/INDEX.md`

## Output

Use the structured contracts in `.claude/rules/12-output-formats.md`.