code-review

name	code-review
description	Code Review Orchestrator. GATE: figma-doctor verifies design spec match first. On PASS, loads 6 sub-skills in parallel based on scope (frontend|backend|full-stack): typecheck-doctor, react-doctor, style-doctor, boundary-doctor, i18n-doctor, backend-doctor. Each sub-skill is independently composable and can be loaded via use_skill. Trigger: code review PR feedback quality check automated scans anti-pattern detection.
version	5

Code Review Orchestrator

Architecture

This skill is the orchestration entry point for code review. All rules live in sub-skills — this skill only coordinates which sub-skills to load and how to aggregate their output.

code-review (this skill)
  │
  ├── GATE: figma-doctor       — Design spec verification (runs FIRST)
  │     ├── ✅ PASS → continue
  │     └── ❌ FAIL → STOP, return to development
  │
  └── PARALLEL (only if gate passes):
        ├── typecheck-doctor    — TypeScript type check
        ├── react-doctor        — React/Redux/Query/Form anti-patterns
        ├── style-doctor        — Styling & design token compliance
        ├── boundary-doctor     — Import boundaries & package dependency direction
        ├── i18n-doctor         — Hardcoded string detection
        └── backend-doctor      — Backend code quality

Each sub-skill is a standalone SKILL.MD containing explicit rules with exact tool invocations to use (search_content, execute_command, read_lints). No shell scripts. Each sub-skill can be loaded individually or composed.

Scope → Skills Mapping

Scope	Gate	Parallel Sub-skills
frontend	figma-doctor	typecheck-doctor, react-doctor, style-doctor, boundary-doctor, i18n-doctor
backend	(skip, no UI)	typecheck-doctor, boundary-doctor, backend-doctor
full-stack	figma-doctor	All 6

Figma gate only applies to UI tasks (frontend / full-stack). Pure backend reviews skip the gate.

Workflow

Step 0: Determine scope & target

Ask the user or infer from changed files what is being built:

• Scope: frontend / backend / full-stack
• Target: what specific feature/section (e.g., "footer", "hero section") — required for figma gate

Step 1 (GATE): Run figma-doctor

If scope includes frontend, run figma-doctor FIRST (sequential). It reads the design spec, finds the matching section, and compares every property against the implementation.

• ✅ PASS → go to Step 2
• ❌ FAIL → STOP. Report all mismatches. Do NOT continue. Return to development.

Step 2: Parallel-load all applicable sub-skills

Use use_skill to load each sub-skill for the determined scope. This MUST be done in parallel — all sub-skill loads in one batch.

Step 3: Execute each sub-skill's rules

Each sub-skill's SKILL.MD lists explicit check rules. Execute them using the tools specified:

• search_content — for regex-based pattern detection
• execute_command — for typecheck, format check, etc.
• read_lints — for compiler/linter diagnostics

All independent checks within and across sub-skills SHOULD run in parallel.

Step 4: Aggregate results

Collect all sub-skill outputs and categorize:

Severity	Meaning	Action
❌ Blocking	Must fix, blocks merge	Fix required
⚠️ Warning	Should fix, does not block	Fix recommended; track as follow-up
💡 Suggestion	Optional improvement	Informational only

Step 5: Output graded report

Present findings organized by sub-skill, with file paths, line numbers, and severity.

Step 6: Deep Review (Optional, for thorough reviews)

After the rule-based scan, manually verify:

1. Architecture — module placement, package layering, component graduation rules
2. Feature — feature completeness, data flow, edge cases, i18n 3-locale sync, a11y
3. Type Contracts — frontend/backend type consistency, API route matching (full-stack only)

Sub-Skill Composition

Each sub-skill is a true CodeBuddy skill and can be:

• Loaded individually via use_skill("code-review/react-doctor")
• Combined on demand — load only the sub-skills needed
• Parallel-loaded by the main agent for full reviews