/review-code                              # Local uncommitted changes (default)
/review-code <file-or-directory>          # Review specific path
/review-code --pr 123                     # Review GitHub PR #123
/review-code --pr 123 --quick             # Quick PR review (Tier 1 only)
/review-code --quick                      # Tier 1 agents only (no multi-LLM)
/review-code --security                   # Security-focused review
/review-code --full                       # All tiers + multi-LLM (most thorough)
/review-code --agents-only                # Skip multi-LLM review
/review-code --llm-only                   # Skip agents, multi-LLM only

/review-code [--pr <number>]
         │
         ▼
┌─────────────────────────────────────────┐
│  Step 1: IDENTIFY CHANGES               │
│  - Local: git diff (uncommitted)        │
│  - PR mode: gh pr diff <number>         │
│  - Detect languages (Go, TS, etc.)      │
│  - Identify domains (API, store, UI)    │
└─────────────────────────────────────────┘
         │
         ▼
┌─────────────────────────────────────────┐
│  Step 2: RUN CODE AGENTS (Claude)       │
│                                         │
│  Tier 1 (always):                       │
│  - race-condition-finder                │
│  - simplicity-reviewer                  │
│  - pattern-reviewer                     │
│  - error-handling-reviewer              │
│                                         │
│  Tier 2 (security):                     │
│  - xss-reviewer                         │
│  - validation-reviewer                  │
│  - permission-auditor                   │
│                                         │
│  Tier 3+ (domain-specific):             │
│  - Based on files changed               │
└─────────────────────────────────────────┘
         │
         ▼
┌─────────────────────────────────────────┐
│  Step 3: MULTI-LLM REVIEW (in parallel) │
│                                         │
│  All models from multi-llm-review.md    │
│  (Codex + Gemini + seq-server)          │
│                                         │
│  Focus: Code quality, edge cases,       │
│  security, performance                  │
└─────────────────────────────────────────┘
         │
         ▼
┌─────────────────────────────────────────┐
│  Step 4: SYNTHESIZE FINDINGS            │
│  - Merge agent + LLM findings           │
│  - Prioritize by severity               │
│  - Apply 80/20 filter                   │
│  - Only 2+ model agreement = MUST FIX   │
└─────────────────────────────────────────┘
         │
         ▼
┌─────────────────────────────────────────┐
│  OUTPUT: Review Report                  │
│  - MUST FIX (blockers)                  │
│  - SHOULD FIX (quality)                 │
│  - Passed checks                        │
└─────────────────────────────────────────┘

Review this code for bugs, security issues, and quality.

## Code Changes
<code>
[paste git diff or file contents]
</code>

## CRITICAL: Apply 80/20 Thinking

**A MUST FIX blocker is ONLY:**
- Bug that will cause runtime failure
- Security vulnerability (injection, auth bypass, XSS)
- Data integrity risk (corruption, loss)
- Race condition / concurrency bug
- Missing error handling that crashes

**NOT a blocker (SHOULD FIX or SKIP):**
- Style issues, naming preferences
- Minor optimizations
- Missing comments/docs
- "Best practices" that don't affect correctness

## Evaluate

1. **Correctness**: Will this code work as intended?
2. **Security**: Any vulnerabilities?
3. **Edge Cases**: Null checks, error handling, boundary conditions?
4. **Performance**: Any obvious inefficiencies?
5. **Patterns**: Does it follow established codebase patterns?

## Output

1. **MUST FIX** (0-3 max): What breaks? File:line? Fix?
2. **SHOULD FIX** (0-5): Quality improvements
3. **VERDICT**: APPROVED / NEEDS WORK

# Pseudo-logic for agent selection
agents = []

# Tier 1: Always run (MUST RUN)
agents.extend([
    "race-condition-finder",
    "simplicity-reviewer",
    "error-handling-reviewer"
])

# Tier 2: Security (always for production code)
if not test_files_only:
    agents.extend([
        "hardcoded-values-reviewer",
        "owasp-security"
    ])

# Tier 3: Backend (Go files)
if has_go_files:
    agents.append("go-pro")
    agents.append("concurrent-go-reviewer")
    if has_db_changes:
        agents.append("postgres-expert")

# Tier 4: Frontend (TS/TSX files)
if has_ts_files:
    agents.append("react-pro")
    agents.append("typescript-pro")

# Tier 5: Testing
if has_test_files:
    if has_go_tests:
        agents.append("test-unit-expert")
    if has_e2e_tests:
        agents.append("playwright-patterns-reviewer")

## Code Review: [files reviewed]

### MUST FIX (Blockers)
| Issue | File:Line | Agent | Fix |
|-------|-----------|-------|-----|
| Race condition in cache access | `cache.go:45` | race-condition-finder | Add mutex |
| Missing permission check | `api.go:123` | permission-auditor | Add HasPermission call |

### SHOULD FIX (Quality)
| Issue | File:Line | Agent | Recommendation |
|-------|-----------|-------|----------------|
| Overly complex function | `utils.go:89` | simplicity-reviewer | Extract helper |

### Passed Checks
- ✅ No XSS vulnerabilities
- ✅ Input validation present
- ✅ Error handling complete
- ✅ Tests cover new code

### Agent Summary
| Agent | Verdict | Findings |
|-------|---------|----------|
| race-condition-finder | ⚠️ ISSUES | 1 race condition |
| simplicity-reviewer | ✅ PASS | - |
| xss-reviewer | ✅ PASS | - |
| permission-auditor | ⚠️ ISSUES | 1 missing check |

---

### Verdict: NEEDS WORK / APPROVED

Fix MUST FIX items before committing.

# Full review of local changes (agents + multi-LLM) - RECOMMENDED
/review-code

# Review specific file
/review-code server/app/item_core.go

# Review a GitHub PR
/review-code --pr 123

# Quick PR review (Tier 1 agents only)
/review-code --pr 123 --quick

# Security-focused PR review
/review-code --pr 123 --security

# Quick review (Tier 1 agents only, no LLM)
/review-code --quick

# Security-focused review
/review-code --security

# Full review (all tiers + multi-LLM)
/review-code --full

# Agents only (skip external LLMs)
/review-code --agents-only

# Multi-LLM only (skip agents)
/review-code --llm-only

/create-plan "feature"     # Create plan
    │
    ▼
/create-code plan.md       # Implement
    │
    ▼
/review-code               # Agent review  ← THIS SKILL
    │
    ▼
Fix MUST FIX items
    │
    ▼
/lint                      # Final formatting
    │
    ▼
Commit changes