repair-domains

Repair Domains Skill

Detect the domain of a repair task and route to the correct skills, review agents, test commands, and fix patterns. A repair can span multiple domains — always identify primary and secondary.

Domain Detection

Run detection in this order. The first match sets the primary domain; subsequent matches become secondary domains.

Step 1: Label-Based Detection (fastest)

If the repair originates from a GitHub issue (#N), read labels:

Label	Domain
area/frontend	frontend
area/backend	backend
area/fullstack	frontend + backend
security	security
devops, ci, infra	devops
priority/critical, priority/urgent	(flag for --hotfix auto-detection, not a domain)

Step 2: File-Path Detection

Search the codebase for files related to the problem statement. Classify by path:

Path pattern	Domain
work-web/src/components/**, work-web/src/app/**	frontend
work-web/src/styles/**, *.css, *.scss, tailwind.*	frontend
work-web/src/hooks/**, work-web/src/contexts/**	frontend
work-agents/src/routes/**, work-agents/src/services/**	backend
work-agents/src/db/**, **/schema.ts, **/migrations/**	backend
work-agents/src/middleware/**	backend (or security if auth-related)
.github/workflows/**, Dockerfile*, docker-compose*	devops
vercel.json, vercel.ts, next.config.*	devops
**/terraform/**, **/pulumi/**, **/cdk/**	cloud-infra
**/auth/**, **/session/**, **/cors.*, **/csp.*	security
work-web/src/lib/api-client.*	frontend (API client layer)
work-agents/src/index.ts	backend (route registration)

Step 3: Content-Based Detection

If file paths are ambiguous, read the problem statement and relevant files for domain signals:

Content signal	Domain
CSS classes, Tailwind utilities, JSX/TSX markup, className	frontend
useState, useEffect, useRouter, component props	frontend
SQL queries, Drizzle ORM, db.select(), db.insert()	backend
API route handlers, Hono, c.json(), c.req.param()	backend
CORS, helmet, csrf, XSS, sanitize, escape	security
JWT, session tokens, bcrypt, argon2, crypto	security
CI pipeline, GitHub Actions YAML, deploy scripts	devops
Cloud provider SDKs, IAM, S3, CloudFront	cloud-infra

Step 4: Default

If no signals detected: backend (most common repair domain in this codebase).

Domain Routing Table

After detection, use this table to configure the repair:

Domain	Skills to load	Review agents	Test command	Fix patterns
frontend	frontend-design:frontend-design	code-reviewer, type-design-analyzer, code-simplifier	cd work-web && bun test	See domain-patterns.md
backend	(none extra)	code-reviewer, silent-failure-hunter, pr-test-analyzer	cd work-agents && bun test	See domain-patterns.md
security	(none extra)	code-reviewer, silent-failure-hunter, pr-test-analyzer	Both test suites + security-specific tests	See domain-patterns.md
devops	vercel:vercel-functions (if Vercel)	code-reviewer	Validate configs + CI lint	See domain-patterns.md
cloud-infra	(none extra)	code-reviewer	terraform validate / pulumi preview	See domain-patterns.md
fullstack	frontend-design:frontend-design	ALL review agents	Both test suites	Combine frontend + backend patterns

Domain-Specific Review Agent Selection

When dispatching review agents during the review-fix phase, select based on domain:

Frontend

- code-reviewer: focus on component logic, prop handling, state management
- type-design-analyzer: only if new types or interfaces are added/modified
- code-simplifier: always — frontend code benefits most from simplification

Backend

- code-reviewer: focus on API correctness, data integrity, query efficiency
- silent-failure-hunter: ALWAYS — backend error handling is critical
- pr-test-analyzer: ALWAYS — backend needs high test coverage

Security

- code-reviewer: focus on auth/authz logic, input validation, output encoding
- silent-failure-hunter: CRITICAL — silenced auth errors are vulnerabilities
- pr-test-analyzer: verify auth edge cases are tested (expired tokens, missing roles, injection)

DevOps

- code-reviewer: focus on config correctness, no hardcoded secrets, proper env var usage

Cloud Infra

- code-reviewer: focus on resource sizing, IAM least-privilege, network rules

Hotfix Auto-Detection

A repair should automatically suggest --hotfix mode when:

1. Issue has label priority/critical or priority/urgent
2. Issue has label security with severity indicators
3. Problem statement mentions "production", "prod", "live site", "customer-facing"
4. The affected files are on the main branch but NOT on staging (rare but possible after a direct main fix)
5. The --url points to a production domain (app.forthai.work, forthai.work, work.forth.ai)

When auto-detected, report: "Auto-detected hotfix: [reason]. Branching from main."

Integration with Repair Command

The repair command should:

1. Run domain detection (Steps 1-4 above)
2. Report detected domain(s): "Domain: frontend (primary), security (secondary)"
3. Load domain-specific skills (if any — use Skill tool)
4. Apply domain-specific search patterns during codebase research
5. Pass domain info to review-fix phase for agent selection
6. Pass domain info to verify phase for appropriate testing