The /meta skill inventories the MetaVibing meta-stack, reviews open friction-ledger entries, identifies gaps, and proposes prioritized improvements for human approval. Use it to audit, review, or check the health of the meta-stack, or invoke it directly as /meta.
Install
npx skillscat add mircus/metavibing/meta Install via the SkillsCat registry.
The /meta skill inventories the MetaVibing meta-stack, reviews open friction-ledger entries, identifies gaps, and proposes prioritized improvements for human approval. Use it to audit, review, or check the health of the meta-stack, or invoke it directly as /meta.
/meta — MetaVibing Meta-Audit
Invoke with: /meta or /meta <specific concern>
This skill runs a meta-audit of the current MetaVibing environment. It surfaces gaps, inconsistencies, and improvement opportunities in the meta-stack.
Procedure
1. Inventory the meta-stack
Check what exists:
CLAUDE.md — present / absent / stale?
.claude/rules/ — what rules exist? any obvious gaps?
.claude/skills/ — what skills exist? any duplicated procedures in CLAUDE.md?
.claude/agents/ — what agents exist? any specialist roles missing?
.claude/hooks/ — what hooks exist? any hard boundaries unprotected?
mcp/ — what tools exist? any manual operations that could be automated?
FRICTION_LEDGER.md — how many open entries? any patterns?
evals/ — any evaluated experiments? any pending?2. Read the friction ledger
For each open entry in FRICTION_LEDGER.md:
- Is the diagnosis complete?
- Has an artifact been created?
- Has an evaluation been run?
- Is it ready to close?
3. Identify gaps
Based on the inventory, identify:
- Procedures described in CLAUDE.md that should be Skills.
- Behavioral rules that are advisory but should be hooks.
- Manual operations repeated more than twice that should be MCP tools.
- Classes of failure in the ledger with no corresponding artifact.
4. Propose improvements
For each gap, propose:
Gap: [description]
Proposed artifact: [type and path]
Priority: high / medium / low
Rationale: [one sentence]Sort by priority. Do not implement — propose for human approval.
5. Summarize
Report:
- Stack health (green / amber / red per layer)
- Open friction entries: N
- Proposed improvements: N
- Recommended next action
Example output
## Meta-Audit — 2026-09-03
### Stack Health
- CLAUDE.md: [OK] present, reasonable scope
- Rules: [OK] 1 rule (taskflow.md), natively loaded from .claude/rules/
- Skills: [OK] 2 skills — /ship-change and /meta, natively loaded from .claude/skills/
- Agents: [OK] 1 agent — final-reviewer, read-only tools enforced via frontmatter
- Hooks: [PLANNED] 0 hooks defined — none implemented yet
- MCP: [PARTIAL] architecture-checker present as a standalone CLI; MCP wrapper is a v1.1 item
- Friction Ledger: [OPEN] N open entries — see FRICTION_LEDGER.md
- Evals: [DESIGNED] baseline charter frozen; 18-trial run not yet executed
### Proposed Improvements (priority order)
1. Run the frozen 18-trial baseline (high) — the central untested claim
2. Build the real MCP wrapper (low) — only once a Friction Ledger entry demands it
### Recommended Next Action
Whatever the Friction Ledger's oldest open entry names, or the 18-trial run if the ledger is clear.