mb-skill-review

skill-review

A composable skill that wraps the Seven Sweeps + auxiliary review gates. /mb-site calls this at pre-lock and pre-publish.

Inputs

• The artifact under review (brief or copy draft)
• The dial (convert | story | brand)
• Reference files (voice.md, audience.md, optional research/*)
• The brief's do_not_state list (for archetype-fidelity check)

Outputs

A synthesized review report:

{
  "ok": false,
  "dial": "convert",
  "sweeps_run": [1, 2, 3, 4, 5, 6, 7],
  "findings": [
    {"sweep": "Specificity", "priority": "P1", "line": 12, "note": "..."},
    ...
  ],
  "panel_score": 7.2,
  "blocking": ["specificity", "ai-tells"]
}

Flow

1. Load the artifact and the dial.
2. Pick sweeps from references/review.md — dial-gated.
3. Spawn one foreground subagent per active sweep, in parallel. Each returns short findings (P1/P2/P3 with line refs).
4. Run auxiliary gates in parallel: De-AI'd, Framework-true, Archetype-fidelity.
5. (Convert dial only) Run Expert Panel scoring across 3-5 personas.
6. Synthesize findings into one report; group by sweep, P-priority, line.
7. Return to caller (which surfaces to operator).

Dial-gated sweeps

Dial	Sweeps
convert	1, 2, 3, 4, 5, 6, 7 + Expert Panel
story	1, 2, 3, 5, 6
brand	1, 2, 6

Cross-references

• references/review.md
• references/anti-patterns.md