Agent Coding Governance
Overview
Two leaves govern agent-assisted development. One sets the gates a machine-written change must pass before merge: the review packet, provenance, licence scan and liability frame. The other covers the repo's instruction file for agents: what goes in it, how long it may run, and when to compact it.
When to use
- Reviewing or accepting agent-written patches; assembling the review packet.
- Provenance, licence-hygiene, liability and authorship questions over generated code.
- Setting up, restructuring or slimming an AGENTS.md or rules file; the agent ignoring repo rules; context-budget problems.
When not to use
- Reviewing a colleague's diff on its merits — outside this skill.
- Designing the tests or the CI gates that enforce the rules — outside this skill.
- Writing human-facing docs, docstrings or API reference — outside this skill.
Precedence
The deepest loaded instruction governs its uniquely-claimed scope; every
scope:string is claimed by exactly one file in this tree; no leaf may widen a prohibition declared by an ancestor — leaves narrow, never override upward.
Subskill index
| Leaf | Task | Trigger |
|---|---|---|
subskills/ai-assisted-coding.md |
Gates, provenance, liability for generated code | "review AI-generated code" |
subskills/agent-context.md |
Repo instruction files for agents | "set up an AGENTS.md" |
See also
subskills/ai-assisted-coding.md— gate the output.subskills/agent-context.md— steer the agent.