instruction-governance-auditor

name	instruction-governance-auditor
description	Use when reviewing, restructuring, creating, removing, or validating AGENTS.md files, Codex skills, installed skill state, or workspace instruction governance so guidance stays coherent with current contracts and runtime behavior.

Instruction Governance Auditor

Use this before changing AGENTS files, skill source, skill registration, or installed skill behavior.

Read First

• docs/instruction-and-skill-governance.md
• workspace-root/ARCHITECTURE.md
• workspace-root/AGENTS.md
• contracts/skills.yaml
• contracts/repos.yaml
• contracts/repo-rules/

Workflow

1. Prove remote freshness for workspace-governance.
2. Inventory active repo AGENTS.md files and registered skill sources.
3. Verify installed managed skills match registered source.
4. Classify each finding as keep, extend, replace, remove, or new.
5. Prefer extending an existing skill or primary operator surface before adding a new skill.
6. Keep AGENTS files short and authority-focused; move repeatable procedure to skills or primary operator docs.
7. If skill source or registration changes, reinstall skills and verify the live install.
8. If workspace-root files change, sync the root and verify the live root.
9. Validate contracts, cross-repo truth, Codex review controls, stale content, and workspace layout before closeout.

Guardrails

• Do not treat source skill changes as complete until live installed skills are refreshed and checked.
• Do not leave stale normal-path instructions for direct Rails, raw OpenProject admin, or raw kubectl exec ... node -e ... ART work.
• Do not create a new skill just because guidance feels broad; first prove the existing skill inventory cannot cover it cleanly.
• Do not bury owner-boundary changes in AGENTS prose only; update the relevant machine-readable contracts too.
• Do not call the workspace clean unless the WGCF clean-state scope passes.