refine-boundaries

Manage the ## Boundaries section in AGENTS.md. Boundaries is the agent autonomy policy — the one section that most commonly grows as the user discovers what agents get wrong, and the one section that benefits from ongoing, error-driven refinement rather than one-shot init.

Philosophy. Every other agent-collabo skill is set-and-forget. Boundaries is the exception. Users almost never know the right rules upfront — they discover them by watching agents make mistakes. This skill turns each mistake into a durable rule, classifies it correctly, and places it where future agents will actually see it before taking the same action.

Scope

In scope:

• Add a new rule to ## Boundaries (Always do / Ask first / Never do)
• Classify a new rule into the correct tier
• Decide whether a rule belongs inline in a topic section (Code Style / Testing / Git Workflow) or cross-cutting in Boundaries
• Restructure a malformed Boundaries section into the 3-tier format
• Create the ## Boundaries section if it does not exist
• Suggest splitting rules out to a dedicated ## Security section when security rules accumulate (≥5)

Out of scope:

• Executing the rules (rules are guidance for the next agent, not runnable checks)