spawn-classifier

File-Path Triggers

Evaluate changed files (from git diff --name-only or the task description) against these patterns:

Changed path pattern	Agent to spawn
*.{go,py,rb,php,java,kt,rs}	backend-developer
app/Http/, app/Services/, src/controllers/, routes/	backend-developer
*.{tsx,jsx,vue,svelte}	frontend-developer
src/components/, src/pages/, resources/js/, app/javascript/	frontend-developer
database/migrations/, db/migrate/, prisma/, *.sql	database-specialist
.github/workflows/, Dockerfile*, docker-compose*.yml	devops-specialist
*.tf, *.tfvars, Pulumi.yaml, serverless.yml	devops-specialist
Jenkinsfile, .gitlab-ci.yml, *.circleci/	devops-specialist
*.spec.{ts,tsx,js}, *.test.{ts,tsx,js}, tests/, spec/ (frontend)	frontend-test-specialist
*.spec.{py,rb,go,java,kt}, tests/, spec/ (backend)	backend-test-specialist
*.stories.{ts,tsx,js}, storybook/, .storybook/	frontend-developer + ui-ux-designer

---

Text / Intent Triggers

Evaluate the task description or user message:

Keyword or phrase	Agent to spawn
"security", "vulnerability", "CVE", "auth bypass", "injection", "XSS", "CSRF"	security-specialist
"penetration test", "threat model", "OWASP"	security-specialist + software-architect
"test", "spec", "coverage", "unit test", "integration test" (frontend context)	frontend-test-specialist
"test", "spec", "coverage", "unit test", "integration test" (backend context)	backend-test-specialist
"design", "UX", "wireframe", "mockup", "user flow", "component library"	ui-ux-designer
"design system", "token", "Figma"	ui-ux-designer + frontend-developer
"SEO", "meta tag", "structured data", "sitemap", "Core Web Vitals", "llms.txt", "GEO", "otimização para IA/LLM"	seo-specialist
Project is a public site, landing page, e-commerce/online store, or blog (see skills/design/seo-optimization/SKILL.md § Detection Signals)	seo-specialist alongside frontend-developer/ui-ux-designer
"database", "schema", "migration", "index", "query", "ORM"	database-specialist
"deploy", "CI/CD", "pipeline", "Docker", "Kubernetes", "infra"	devops-specialist
"architecture", "ADR", "design decision", "trade-off", "system design"	software-architect
"API", "REST", "GraphQL", "endpoint", "service"	backend-developer
"bug", "fix", "regression", "broken", "error"	context-dependent — see Default below

---

Multi-Agent Combinations

Some tasks require more than one agent. Spawn in parallel when there are no dependencies:

Scenario	Agents
Full-stack feature	backend-developer + frontend-developer + optionally database-specialist
Security audit	security-specialist + software-architect
New API endpoint with UI	backend-developer → frontend-developer (sequential — API first)
Database schema + migration	database-specialist → backend-developer (sequential — schema first)
Test-only task (both layers)	backend-test-specialist + frontend-test-specialist (parallel)
PR review	code-reviewer + security-specialist + optionally database-specialist

---

Conflict Resolution

When a changed file matches multiple patterns:

1. Most specific pattern wins. src/controllers/UsersController.ts matches both frontend (src/) and backend (controller) — the controllers/ pattern is more specific, so spawn backend-developer.
2. When genuinely ambiguous, spawn software-architect first to classify and produce a plan.
3. Never spawn an agent for a domain not touched by the task — even if the agent is "usually helpful".

---

Default Rule

When scope is unclear — no strong file-path or text match — ask before spawning, per the Quiz-first Rule (skills/shared/interaction-patterns/SKILL.md):

1. Use AskUserQuestion (single-select) offering the likely domains: Backend, Frontend, Database, DevOps, Security, Architecture, and "Not sure — let an agent classify it."
2. If the user picks a domain, spawn the matching agent directly — no classification agent needed.
3. Only if the user picks "Not sure" does this fall back to spawning software-architect first to analyze the task and produce a plan that identifies which domains are affected; downstream agents are then spawned from that plan.

This avoids paying for a full reasoning-tier agent invocation just to classify an ambiguous request, while still preventing incorrect agents from producing work that conflicts with the actual architecture when the user genuinely doesn't know.

---

Classification Checklist

• Run git diff --name-only HEAD~1 to get changed files
• Match files against path-pattern table above
• Match task description against text-trigger table above
• Identify parallel vs. sequential spawn order
• If ambiguous, spawn software-architect first