메뉴
Use when dispatching code reviews, spec reviews, or PR reviews. Defines the 5 mandatory specialized reviewers, their domains, dispatch rules, and the PASS/FAIL cycle. Also lists all 12 agents and their configurations.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Check changelogs for Bevy, Next.js, Zustand, wasm-bindgen, Rapier, and all SpawnForge deps. Use when training data may be stale, before major upgrades, or when a library API behaves unexpectedly.
Forward-looking sister to changelog-review. Scans our SaaS + library stack (analytics, monitoring, observability, agent tracking, infrastructure, AI, payments, auth) against TODAY's provider changelogs / releases / betas to surface capabilities we are NOT yet using — newly shipped or long-available-but-unadopted — that could improve the product or our build harness. Every opportunity comes with its provider cost. Use for quarterly capability audits, when asking "what new features can we leverage," or before planning a growth/observability/infra initiative.
SpawnForge Stripe webhook patterns — idempotent refunds, CTE atomic claims, neonSql transaction ordering, token billing. Auto-loads when editing billing code.
Monitor open PRs for CI failures and review comments, then fix issues automatically. Use when PRs need to be driven to green — checks CI status, Sentry comments, merge conflicts, and applies fixes.
Review a PR for SpawnForge bugs, security issues, and regressions. Posts inline GitHub comments with severity markers. Use when asked to "review PR", "code review", or "check this PR". Invoke with /pr-code-review <PR number>.
Taskboard management skill. View, create, update, and manage tickets on the project taskboard. Enforces standard ticket format with user stories and acceptance criteria.
| name | review-protocol |
| description | Use when dispatching code reviews, spec reviews, or PR reviews. Defines the 5 mandatory specialized reviewers, their domains, dispatch rules, and the PASS/FAIL cycle. Also lists all 12 agents and their configurations. |
All specs, plans, and PRs go through 5 antagonistic specialized reviewers. Reviews are PASS or FAIL only — any issue at any severity is a FAIL. Loop until all 5 pass clean.
| Role | Agent Type | Focus |
|---|---|---|
| Architect | feature-dev:code-architect | Structure, dependencies, scaling, monorepo, build pipeline |
| Security | security-reviewer | Injection, auth, data exposure, validation, XSS, CSRF |
| DX | dx-guardian | Developer workflow, onboarding, migration burden, documentation |
| UX/Frontend | ux-reviewer | Accessibility (WCAG AA), component UX, Tailwind, responsive |
| Test | test-writer | Coverage gaps, CI gates, parameterization, visual regression |
code-reviewer for the 5 specialized agentsinfra-devopsdocs-guardian (PASS/FAIL only).claude/agents/ — 12 agents)All agents have: memory, effort, model, tools, skills, and agent-scoped hooks in frontmatter.
| Agent | Key Config | Trigger |
|---|---|---|
builder | isolation: worktree, memory: user | Implementation tasks |
validator | mcpServers: playwright | QA gate, validation suite |
planner | model: opus, memory: user | Architecture, specs |
docs-guardian | background: true, read-only | Doc review (PASS/FAIL) |
dx-guardian | background: true, model: haiku | DX audits |
security-reviewer | background: true, read-only | Security audits |
test-writer | memory: project | Vitest + RTL tests |
infra-devops | mcpServers: github | Deploy, CI/CD |
ux-reviewer | background: true, mcpServers: playwright | UX/a11y |
code-reviewer | background: true, read-only | PR review |
docs-maintainer | memory: project | Documentation |
rust-engine | mcpServers: context7 | Bevy ECS, WASM |
All 5 reviewers have: background: true, read-only tools, Stop hook validates PASS/FAIL, PreToolUse blocks writes.
Agent teams: Enabled via CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1 in settings.json.