Agent System & Skills Matrix
Universal Squad Workflow (7 Steps)
Step 1: RESEARCH (all agents parallel)
Read relevant files, load SKILL.md, WebSearch best practices
Step 2: INDIVIDUAL PLANS (written, not code)
What found, what to change, why, risks, acceptance criteria
Step 3: TEAM LEAD PLAN REVIEW (gate)
Reject insufficient research, resolve conflicts, approve with priority
Step 4: IMPLEMENTATION (parallel, file-isolated)
Only approved changes, no overlaps
Step 5: DUAL REVIEW
Agent 2 (Security) + Agent 5 (Quality) review ALL changes
Step 6: TEAM LEAD FINAL
Read every file, run pnpm check + test + build
Step 7: DOCUMENT LEARNINGS
Add to .claude/rules/remember.md
Extended Specialist Pool
| Agent | Focus |
|---|
| Database Architect | Schema, SQL optimization |
| Test Engineer | Testing strategies |
| Debugger | Root cause analysis |
| Performance Optimizer | Core Web Vitals |
| SEO Specialist | Ranking, visibility |
| DevOps Engineer | CI/CD, Docker |
| AI Engineer | AI/ML, prompt engineering |
| Explorer Agent | Codebase analysis |
| Penetration Tester | Offensive security |
Skills Decision Matrix
| Task Keywords | Load Skills |
|---|
| component, UI, page | react-patterns, tailwind-patterns, frontend-design |
| API, endpoint, tRPC | api-patterns, nodejs-best-practices |
| auth, login, session | validatestrategy-auth, vulnerability-scanner |
| payment, stripe, webhook | validatestrategy-payment-flow, vulnerability-scanner |
| analysis, pipeline, AI | validatestrategy-analysis-pipeline, performance-profiling |
| email, Resend, nurture | validatestrategy-email-system |
| admin, dashboard | validatestrategy-admin-monitoring, react-patterns |
| database, schema, Drizzle | database-design |
| test, spec, TDD | testing-patterns, tdd-workflow, webapp-testing |
| security, OWASP | vulnerability-scanner, red-team-tactics |
| SEO, meta, sitemap | seo-fundamentals, bot-detection-seo |
| render, prerender, SSR | rendering-mastery, prerendering-infrastructure |
| style, CSS, animation | tailwind-patterns, frontend-design |
| performance, Lighthouse | performance-profiling, rendering-mastery |
| deploy, CI/CD | deployment-procedures, docker-expert |
| debug, bug, error | systematic-debugging, testing-patterns |
| content, blog, MDX | seo-content, seo-fundamentals |
Skill Combo Recipes
| Scenario | Skills |
|---|
| New Full-Stack Feature | brainstorming + plan-writing + react-patterns + api-patterns + database-design + testing-patterns + vulnerability-scanner |
| Security Audit | vulnerability-scanner + red-team-tactics + code-review-checklist |
| Performance Sprint | performance-profiling + rendering-mastery + tailwind-patterns |
| SEO Pipeline | seo-fundamentals + seo-content + seo-schema + bot-detection-seo |
| Bug Investigation | systematic-debugging + testing-patterns + domain skill |
| Auth Change | validatestrategy-auth + vulnerability-scanner + red-team-tactics |
Project-Specific Skills (5 Critical)
| Skill | Domain | When |
|---|
validatestrategy-payment-flow | Payments | Webhook, Stripe, PayPal, crypto |
validatestrategy-auth | Auth | SIWE, magic link, JWT, nonces |
validatestrategy-analysis-pipeline | AI Analysis | 6-part, state machine, prompts |
validatestrategy-email-system | Email | Resend, nurture, rate limiting |
validatestrategy-admin-monitoring | Admin | Event sourcing, circuit breaker |
Mandatory Review Protocol
NEVER declare work complete without launching 3-4 parallel review agents to check:
- Every changed file for logic errors
- All intended changes are present
- No contradictions between files
- SEO/security/performance implications
Agent Prompt Template
You are [ROLE]. Your mandate is [MANDATE].
Project Brief: [BRIEF]
Task: [SPECIFIC TASK]
Output: Clear sections, concrete recommendations, flag risks, list assumptions.