| name | liva-skill-governance |
| description | Audit and maintain LIVA's Claude, Codex, and Obsidian skill knowledge. Use when adding, updating, or removing a skill; checking skill frontmatter or metadata; investigating stale or foreign instructions; validating cross-agent parity; or preparing skill changes for review. |
LIVA Skill Governance
Workflow
- Read
AGENTS.md, CLAUDE.md, and git status --short. Preserve every pre-existing dirty file.
- Call the Obsidian
search_vault tool for skill workflow coding standards testing guidelines, then read the matched sources. Do not replace this step with text search while the MCP server is available.
- Run
npm run skills:audit. Errors block completion; warnings are technical debt that must be classified.
- Scaffold new skills with the official
skill-creator scripts. A SKILL.md frontmatter contains only name and description; agent UI and tool dependencies belong in agents/openai.yaml.
- When a skill exists in both
.claude/skills/ and .agents/skills/, keep the two directories byte-identical. Vault notes should point to the canonical skill instead of duplicating its full instructions.
- Validate every changed skill with the official
quick_validate.py, run npm run test:skills, then rerun npm run skills:audit.
- Report scanned counts, errors, warnings, and exact paths. Follow the repository Git boundary: stage only the authorized files and leave commit, merge, and remote operations to the user.
Metadata dialects
SKILL.md: name, description.
agents/openai.yaml: interface, optional tool dependencies, and invocation policy.
- Obsidian Vault:
title, tags, author, last_update.
Never copy keys between these dialects merely to satisfy a validator.
Stop conditions
Stop and report the evidence when search_vault is unavailable, an audit error remains, a scaffold placeholder remains, a relative link is broken, or mirrored skill directories differ.