Run Phase 3 (adversarial review) for the active VCSDD feature. Spawns a fresh vcsdd-adversary agent with zero Builder context to review implementation against spec. Produces binary PASS/FAIL verdict per dimension.
Scanned 6/1/2026
Install via CLI
openskills install sc30gsw/vcsdd-claude-code---
name: vcsdd-adversary
description: Run Phase 3 (adversarial review) for the active VCSDD feature. Spawns a fresh vcsdd-adversary agent with zero Builder context to review implementation against spec. Produces binary PASS/FAIL verdict per dimension.
---
## What
Runs the adversarial review (Phase 3). Spawns a fresh vcsdd-adversary agent (opus model, review-output-only, zero Builder context) to review the implementation against the spec across 5 dimensions.
## When
Run after `/vcsdd-impl` completes Phases 2b and 2c. Requires active feature at phase `2c`. In strict mode, `/vcsdd-contract-review` must already have produced a PASS verdict for the current sprint contract.
## How
1. **Resolve sprint number** from `state.json.sprintCount` (created when Phase 2a starts)
2. **In strict mode, validate contract gate** before review:
- `contracts/sprint-N.md` must exist
- frontmatter must include `status: approved`
- the contract must define at least one `CRIT-XXX` criterion
- `reviews/contracts/sprint-N/output/verdict.json` must exist and `overallVerdict` must be `PASS`
- `reviewContext.contractPath` must reference `contracts/sprint-N.md`
- `reviewContext.contractDigest` must still match the approved contract snapshot
- `iteration` must equal `negotiationRound + 1`
3. **Write review manifest** to `reviews/sprint-N/input/manifest.json`:
```json
{
"reviewType": "implementation",
"featureName": "...",
"sprintNumber": N,
"contractPath": "contracts/sprint-N.md",
"artifactsToReview": {
"spec": ["specs/behavioral-spec.md", "specs/verification-architecture.md"],
"tests": ["tests/..."],
"source": ["src/..."]
},
"reviewDimensions": ["spec_fidelity", "edge_case_coverage", "implementation_correctness", "structural_integrity", "verification_readiness"]
}
```
4. **Create output directories**: `reviews/sprint-N/output/findings/`
5. **Spawn FRESH vcsdd-adversary agent**: this MUST be a new agent with no Builder context
6. **Collect outputs** after adversary completes:
- `reviews/sprint-N/output/verdict.json`
- `reviews/sprint-N/output/findings/FIND-NNN.json` (one per finding)
7. **Record gate**: `recordGate(feature, '3', overallVerdict, 'adversary')`
8. **If PASS**: transition to Phase 5 in all modes, display summary
9. **If FAIL**: display findings grouped by dimension/category, proceed to `/vcsdd-feedback`
## Fresh Context Requirement
The adversary MUST be spawned as a new Agent instance. Do NOT:
- Pass Builder conversation history
- Share the current conversation context
- Let the adversary see Builder's reasoning
The adversary reads ONLY from disk (review manifest + source files).
## Examples
```bash
/vcsdd-adversary
/vcsdd-adversary --sprint 2 # specify sprint number explicitly
```

No comments yet. Be the first to comment!