Qcritical-review

Qcritical-review — Adversarial Verification

Role

Stress-tests artifacts at each SIVS stage through adversarial sub-agents. Produces a structured PASS/WARN/FAIL verdict. Designed to be called standalone or auto-invoked by other SIVS skills.

CLI Interface

/Qcritical-review --stage spec                  # Review a spec document
/Qcritical-review --stage verify                # Review an implementation
/Qcritical-review --stage supervise             # Review merge readiness
/Qcritical-review --mode cross-model            # Use both Claude + Codex as reviewers
/Qcritical-review --stage verify --mode cross-model   # Combine stage + mode
/Qcritical-review <file>                        # Auto-detect stage from file type
/Qcritical-review                               # Auto-detect from recent SIVS context

Review Modes

Mode	Agents	When to Use
claude-only (default)	3 Claude sub-agents	Fast, low-cost reviews
cross-model	2 Claude + 1 Codex	High-stakes reviews needing independent model perspectives

In cross-model mode, the most adversarial agent per stage is routed to Codex:

• Spec: Edge Case Finder → Codex
• Verify: Devil's Advocate → Codex
• Supervise: Merge Blocker → Codex

This ensures the strongest critic uses a genuinely independent model, eliminating same-model confirmation bias.

9-Step Protocol

Each review session runs a structured sequence of up to 9 steps drawn from the OMC critic protocol. Not every SIVS stage runs all 9 steps — see the stage mapping column in the Stage Detection table below.

#	Step	Summary
1	Pre-commitment Prediction	Before reading, commit to 3–5 predicted problem areas
2	Multi-perspective Review	Examine through SE / Junior / Ops lenses in parallel
3	Pre-Mortem	Generate 5–7 failure scenarios assuming exact execution
4	Ambiguity Scan	Identify steps with two valid but conflicting interpretations
5	Devil's Advocate	Argue the implementation is wrong; hunt for crashes and silent failures
6	Self-audit	Re-examine each CRITICAL/MAJOR finding for confidence and bias
7	Realist Check	Pressure-test severity against realistic worst-case and mitigations
8	Adversarial Escalation	Trigger max-adversarial mode on CRITICAL findings or 3+ MAJOR
9	Explicit Gap Analysis	Catalog what is missing — requirements, assumptions, omitted context

Full definitions (trigger conditions, output schemas, examples): ./reference/nine-step-protocol.md

Stage Detection (when --stage is omitted)

Stage	Detected From	9-Step Mapping
spec	TASK_REQUEST*.md or spec file	Steps 1, 2, 4, 9 (Pre-commitment, Multi-perspective, Ambiguity Scan, Gap Analysis)
verify	Source code or diff	Steps 3, 5, 6 (Pre-Mortem, Devil's Advocate, Self-audit)
supervise	PR or merge context	Steps 7, 8 (Realist Check, Adversarial Escalation)

Detection order:

1. If a file argument is given: match against the Stage column above
2. If no argument: check .qe/state/unified-state.json for last SIVS stage
3. If ambiguous: ask user via AskUserQuestion

Execution Procedure

Step 1: Gather Target Artifact

Stage	What to Read
spec	TASK_REQUEST file, VERIFY_CHECKLIST, any referenced design docs
verify	git diff of implementation, test results, checklist status
supervise	Full PR diff (git diff main...HEAD), CI status, review comments

Step 2: Spawn Adversarial Agents

Spawn 3 sub-agents in parallel via the Agent tool. Each adopts a distinct critical lens. Agents must NOT see each other's output.

Spec Stage Agents

The Spec stage runs two cognitive modes — Structural (구조적 사고) and Critical (비판적 사고) — plus a boundary-focused finder. Full mode definitions (posture, key questions, adversarial instruction, must-nots) live in ./reference/thinking-modes.md. These agents implement the mandatory Spec self-reference gate — see ./reference/spec-gate-protocol.md.

Agent	Mode	Role	Key Questions
Structural Reviewer	Structural	Stress-test the spec's structure for completeness & internal consistency	"Does every goal map to an item and vice versa? Any contradictory requirements? Dangling dependencies? Subjective/unverifiable items? Whole sub-problems missing?"
Critical Reviewer	Critical	Devil's advocate on the spec's substance	"What false assumption is this built on? What error case / production scenario is absent? Where will this spec lead the implementer wrong?"
Edge Case Finder	Critical (boundary)	Identify boundary conditions	"What happens at zero? At max? With concurrent access? With malformed input? With network failure?"

The Critical Reviewer is the designated most-adversarial agent and is the one auto-upgraded to a cross-model engine when codex is reachable (see Engine Routing per Mode below).

Verify Stage Agents

Cognitive mode: Critical (비판적 사고) — see ./reference/thinking-modes.md Mode 2. These three agents implement the mandatory Verify gate — ./reference/verify-gate-protocol.md. Devil's Advocate is the cross-model-upgrade target.

Agent	Role	Key Questions
Devil's Advocate	Argue the implementation is wrong	"Where does this break? What input crashes it? Which test is missing?"
Security Auditor	Find vulnerabilities	"Is there injection? Auth bypass? Data leak? OWASP Top 10 exposure?"
Performance Skeptic	Challenge efficiency	"What's the time complexity? Does it scale? Are there N+1 queries? Memory leaks?"

Supervise Stage Agents

Cognitive mode: Meticulous (꼼꼼한 사고) — see ./reference/thinking-modes.md Mode 3. These three agents implement the mandatory Supervise gate (runs only after binary Verify passes) — ./reference/supervise-gate-protocol.md. Merge Blocker is the cross-model-upgrade target.

Agent	Role	Key Questions
Merge Blocker	Argue against merging	"What regression risk exists? Is test coverage sufficient? Are there unresolved TODOs?"
Merge Advocate	Argue for merging	"What's the cost of delay? Is the remaining risk acceptable? Does it meet the spec?"
Impartial Judge	Weigh both sides	"Which concerns are valid? Which are hypothetical? What's the actual risk level?"

Step 3: Each Agent Output Format

Each agent MUST return a structured analysis:

## [Agent Role]

### Findings
1. [Finding with severity: CRITICAL / HIGH / MEDIUM / LOW]
2. ...

### Evidence
- [Specific file:line or section reference for each finding]

### Verdict: [PASS | WARN | FAIL]
- FAIL: Found critical or high-severity issues that must be addressed
- WARN: Found medium issues worth discussing
- PASS: No significant concerns from this perspective

Step 4: Aggregate Verdicts

Collect all 3 agent reports and produce a unified verdict:

Critical Review Report
══════════════════════

Stage: [spec | verify | supervise]
Target: [artifact name/path]

┌─ Gap Hunter ─────────────── WARN ─┐
│ 2 medium findings                  │
│ - Missing error handling for X     │
│ - No mention of concurrent access  │
└────────────────────────────────────┘

┌─ Scope Critic ───────────── PASS ─┐
│ No significant concerns            │
└────────────────────────────────────┘

┌─ Edge Case Finder ───────── FAIL ─┐
│ 1 critical finding                 │
│ - Division by zero when count = 0  │
└────────────────────────────────────┘

Overall: FAIL
Reason: 1 critical finding requires resolution before proceeding.

Action Items:
  1. [CRITICAL] Handle division by zero in calculate_average()
  2. [MEDIUM] Add error handling for timeout scenario
  3. [MEDIUM] Document concurrent access behavior

Step 5: Verdict Rules

Condition	Overall Verdict
Any agent returns FAIL	FAIL
2+ agents return WARN	WARN
1 agent returns WARN, rest PASS	PASS (with notes)
All agents return PASS	PASS

Step 6: Present to User

Display the full report, then ask:

• On FAIL: "Address the critical items before proceeding. Want me to fix them?"
• On WARN: "Review the warnings. Proceed anyway or address them first?"
• On PASS: "No critical issues found. Proceed to next stage."

Agent Spawn Rules

1. All 3 agents run in parallel (single message, 3 Agent tool calls)
2. Agent prompts must include:
   • The full artifact content (spec text, diff, or PR summary)
   • Their assigned role and questions (from the stage table above)
   • The required output format
   • Instruction: "Be adversarial. Your job is to find problems, not confirm quality."
3. Agents must NOT be told what other agents are looking for

Engine Routing per Mode

claude-only (default):

• All 3 agents use subagent_type: "general-purpose"

cross-model:

1. First, check Codex availability:

   node -e "(async()=>{const {pathToFileURL}=await import('url');const {join}=await import('path');const base=process.env.CLAUDE_PLUGIN_ROOT||join(process.env.HOME||process.env.USERPROFILE||'','.claude');const m=await import(pathToFileURL(join(base,'scripts','lib','codex_bridge.mjs')).href);const r=await m.getCodexPluginInfo();console.log(JSON.stringify(r))})()"
   

2. If installed: true: route the designated adversarial agent to Codex via subagent_type: "codex:codex-rescue"
3. If installed: false: fall back to claude-only mode with a notice

Stage	Codex Agent	Why This One
spec	Critical Reviewer	The strongest spec critic should be a genuinely different engine
verify	Devil's Advocate	The strongest critic should be a different model
supervise	Merge Blocker	Merge opposition must be genuinely independent

The remaining 2 agents always use Claude sub-agents.

Automatic cross-model upgrade (mandatory Spec gate)

The manual --mode cross-model above is opt-in. The mandatory Spec self-reference gate (invoked by Qgenerate-spec Step 2.6) instead upgrades automatically and with zero configuration (DECISION_LOG D012):

1. Baseline (always runs): all Spec agents are same-engine sub-agents (subagent_type: "general-purpose"). Fully functional with no codex installed — independence comes from fresh context + adversarial role.
2. Auto-upgrade (best-effort): detect codex reachability via getCodexPluginInfo() / isCodexReachable() from scripts/lib/codex_bridge.mjs. If reachable, route the Critical Reviewer to subagent_type: "codex:codex-rescue" for a truly independent engine.
3. Graceful degrade: if codex is absent or unreachable, silently keep the same-engine baseline. Codex is never a required dependency.

This makes the strongest critic genuinely independent when possible, while guaranteeing the gate always runs even in an all-Claude (or all-Codex) homogeneous setup — which is exactly the self-reference case this gate exists to defend.

The same automatic upgrade applies to the Verify gate (cross-model target = Devil's Advocate) and the Supervise gate (cross-model target = Merge Blocker).

Cross-model failure fallback (all gates): a best-effort upgrade must never block a mandatory gate or silently pass as if it were cross-model.

• If the codex sub-agent errors or times out → log crossmodel=false + reason, re-run that one agent on Claude (general-purpose), and mark the gate result degraded → at least WARN (independence was reduced).
• If the Claude re-run also fails → WARN-blocked (NOT PASS), requiring explicit user override, with audit reason=double-failure.

Bootstrap clause (reviewing Qcritical-review itself)

When a change touches Qcritical-review or its reference/*-gate-protocol.md files, the gate cannot trust its own (possibly-changed) behavior to review that change — a self-reference within the self-reference defense. In that case the review MUST run against the pre-change baseline of these files plus an explicit diff inspection of the proposed change, rather than the in-tree (modified) version. This prevents a broken gate edit from approving itself.

Report Labeling

In cross-model mode, each agent box in the report shows the engine used:

┌─ Devil's Advocate [Codex] ─── FAIL ─┐
┌─ Security Auditor [Claude] ── WARN ─┐
┌─ Performance Skeptic [Claude]─ PASS ─┐

Integration Points

This skill is designed to be called by other SIVS skills:

Caller Skill	When	Stage
Qgenerate-spec / Qgs	After spec generation	spec
Qcode-run-task	After verify loop passes	verify
Esupervision-orchestrator	Before final verdict	supervise

Callers invoke via: /Qcritical-review --stage <stage>

Will

• Spawn 3 adversarial sub-agents per stage
• Produce structured PASS/WARN/FAIL verdict with evidence
• Run agents in parallel for speed
• Adapt critical lens to SIVS stage

Will Not

• Replace Qdebate for open-ended topic debates
• Replace Qperspective for general multi-viewpoint analysis
• Auto-fix issues (only identify and report)
• Run more than 3 agents (focused critique over broad coverage)

Attribution

9-Step Protocol adapted from oh-my-claudecode (MIT, © 2025 Yeachan Heo): https://github.com/Yeachan-Heo/oh-my-claudecode/blob/main/agents/critic.md