agent-capability-map

Agent Capability Map Reference

> **Scope**: Task-to-agent routing table; capabilities and boundaries per agent.

---

Primary Routing Table

| Task Type | Primary Agent | Fallback | Avoid | |-----------|--------------|----------|-------| | Go source code | `golang-general-engineer` | `golang-general-engineer-compact` | general-purpose | | TypeScript backend API | `nodejs-api-engineer` | `typescript-frontend-engineer` | general-purpose | | TypeScript frontend/React | `typescript-frontend-engineer` | — | nodejs-api-engineer | | Python scripts/data | `python-general-engineer` | `python-openstack-engineer` | general-purpose | | Database schema/migrations | `database-engineer` | — | application agents | | Kubernetes/Helm | `kubernetes-helm-engineer` | — | ansible-automation-engineer | | Ansible playbooks | `ansible-automation-engineer` | — | kubernetes-helm-engineer | | OpenSearch/Elasticsearch | `opensearch-elasticsearch-engineer` | — | database-engineer | | Prometheus/Grafana | `prometheus-grafana-engineer` | — | — | | Swift iOS/macOS | `swift-general-engineer` | — | — | | Kotlin Android/JVM | `kotlin-general-engineer` | — | — | | PHP backend | `php-general-engineer` | — | — | | React Native / Expo | `react-native-engineer` | — | typescript-frontend-engineer | | Code review | `reviewer-code-playbook` | `reviewer-system-playbook` | — | | Security audit | `security-threat-model` | — | — | | Performance | `performance-optimization-engineer` | — | — | | Documentation | `technical-documentation-engineer-playbook` | — | — |

Verify agent exists: `ls ~/.claude/agents/ | grep {agent-name}`

---

Agent Scope Boundaries

**golang-general-engineer**: CAN modify `.go` files, run `go build/test/vet`, add packages, fix linting. CANNOT modify non-Go files, make architectural decisions, run migrations.

**nodejs-api-engineer**: CAN do REST endpoints, middleware, auth, webhooks, DB integration. CANNOT do frontend components, native code, infrastructure.

**database-engineer**: CAN do schema design, migrations, indexes, query optimization. CANNOT do application code. **Mandatory sequencing**: database agent ALWAYS before application agents on schema changes.

**performance-optimization-engineer**: CAN do Core Web Vitals, bundle analysis, rendering optimization, profiling. CANNOT fix unrelated logic bugs.

---

Compound Task Routing

API + Frontend Feature

1. database-engineer → schema migration
2. nodejs-api-engineer → endpoint (parallel with 3 if no shared types)
3. typescript-frontend-engineer → UI
4. reviewer-code-playbook → full-stack review

Go Service Performance

1. performance-optimization-engineer → profile, identify hotspot (read-only)
2. golang-general-engineer → fix
3. golang-general-engineer → benchmark validation

Infrastructure + Application

1. kubernetes-helm-engineer → Helm chart (parallel with 2 if isolated)
2. ansible-automation-engineer → provisioning
3. Application agent → smoke test

---

Preferred Patterns

Route to Specialist Agent

**Signal**: Dispatching undefined subagent_type for Go errors. **Fix**: Look up task type in routing table, specify exact `subagent_type`.

Match Agent to Language Domain

**Signal**: TypeScript backend work sent to frontend engineer. **Fix**: Check imports — `express`/`@nestjs`/`fastify` → nodejs-api-engineer. `react`/`next` → typescript-frontend-engineer.

rg "import.*express|import.*@nestjs" --files-with-matches src/  # → backend
rg "import.*react|import.*next" --files-with-matches src/       # → frontend

Route Schema Changes to Database Agent

**Signal**: Asking nodejs-api-engineer to add a column. **Fix**: database-engineer first for migration, then application agent for ORM models.

---

Agent Selection Decision Tree

.go files? → golang-general-engineer
TypeScript backend/API? → nodejs-api-engineer
TypeScript frontend/React? → typescript-frontend-engineer
Database schema? → database-engineer (first, before app agents)
K8s/containers? → kubernetes-helm-engineer
VM provisioning? → ansible-automation-engineer
Performance? → performance-optimization-engineer
Code review? → reviewer-code-playbook
Otherwise? → Check agents/ for domain match → general-purpose as last resort