| name | review-adversarial |
| description | Challenge a design, plan, or artifact by actively looking for weaknesses, risks, and blind spots. Use when the user says 'challenge this design', 'red team this', 'steelman the alternative', or 'find what we missed'. |
Adversarial Review Workflow
Goal
Act as a rigorous challenger — not a critic for its own sake, but a structured devil's advocate. Find the real risks, unspoken assumptions, and failure modes that optimistic reviews miss.
Step 1: Load the Artifact
Ask: What are we reviewing?
- PRD / product brief
- Architecture design / ADR
- Epic or story set
- Migration plan
- Any other document or decision
Read the artifact in full before proceeding.
Step 2: Assumption Mapping
List every explicit and implicit assumption the artifact makes:
| # | Assumption | Evidence for it | What if it's wrong? |
|---|
| 1 | ... | ... | ... |
Flag any assumption with no evidence as HIGH RISK.
Step 3: Failure Mode Analysis
For each major decision or component, ask:
- What is the most likely way this fails? (not the most dramatic — the most probable)
- What external dependency could break this?
- What happens under 10x the expected load / scale / users?
- What does an adversary (competitor, attacker, regulator) do to break this?
Produce a table:
| Component / Decision | Failure Mode | Probability (H/M/L) | Impact (H/M/L) | Mitigation |
|---|
| ... | ... | ... | ... | ... |
Step 4: Steelman the Strongest Alternative
Identify the best alternative approach that was NOT chosen.
Argue for it as strongly as possible:
- Why it would work better
- What evidence or assumptions favour it
- What the current choice loses by not going this way
This is not a recommendation to switch — it surfaces the real trade-off.
Step 5: Blind Spot Scan
Ask the questions no one wants to ask:
- What is this design completely silent on? (security, cost, ops burden, rollback)
- Who has a stake in this failing? (internal politics, competing teams, vendors)
- What regulatory or compliance angle was ignored?
- What would a new team member find confusing in 6 months?
Step 6: Output
Produce an adversarial review report:
## Adversarial Review — {artifact title}
### Critical Assumptions (unsupported)
- ...
### Top 3 Failure Modes
1. [component] — [failure] — [mitigation]
2. ...
3. ...
### Strongest Alternative (steelmanned)
...
### Blind Spots
- ...
### Recommended Actions Before Proceeding
1. ...
2. ...
Save to {docsPath}/analysis/adversarial-review-{slug}.md.
