| name | multi-model-adversarial-review |
| description | Use for high-stakes review where you want multiple model providers reviewing the same artifact and critiquing each other. Shells out to installed coding-agent CLIs (claude, codex, gemini, pi, opencode by default; amp and droid available opt-in) to run parallel reviews, then runs a cross-critique grid where each reviewer evaluates the others' findings to catch hallucinations and severity inflation, then synthesizes a final deduplicated report. Triggers on "MMAR review", "multi-model review", "cross-model adversarial", "review with all the models", or when single-model PAR feels insufficient. |
Multi-Model Adversarial Review (MMAR)
A three-stage review pipeline that uses multiple installed coding-agent CLIs as independent reviewers, then has them critique each other's findings, then synthesizes a final report. Catches model-specific blind spots and hallucinations that single-model PAR cannot.
When To Use This vs. Plain PAR
| Situation | Use |
|---|
| Routine review, normal stakes | parallel-adversarial-review (faster, cheaper) |
| Pre-merge review on hot path code | MMAR |
| Security review | MMAR |
| Production incident postmortem code change | MMAR |
| You suspect a model has a blind spot for this kind of code | MMAR |
| Compliance / audit artifact | MMAR |
MMAR costs more (N+1 model invocations + cross-critique). Don't reach for it on every commit.
Pipeline
Stage 1: Parallel Reviews
┌──────────────┬──────────────┬──────────────┐
diff ──────► │ claude │ codex │ gemini │ ───► findings_<model>.md
└──────┬───────┴──────┬───────┴──────┬───────┘
│ │ │
Stage 2: Cross-Critique (NxN-1 grid)
┌──────────────┬──────────────┬──────────────┐
│ codex critiq │ gemini critiq│ codex critiq │
│ of claude │ of claude │ of gemini │ ───► critique_<a>_of_<b>.md
│ ... │
└──────┬───────┴──────┬───────┴──────┬───────┘
│ │ │
Stage 3: Synthesis
┌──────────────────────────────────────────────┐
│ synthesizer (Claude as subagent or CLI) │
│ - dedupe across reviewers │
│ - drop hallucinations flagged by critics │
│ - apply severity-disagreement rule │
│ - produce final findings report │
└──────────────────────────────────────────────┘
