Agents

Lean reference for the built-in Claude subagents.

Why these agents

Clear separation of roles so you can pick the right depth quickly.
Safer behavior by default through least-privilege tools and settings guardrails.
Faster handoffs across planning, implementation, review, and docs.

Agent Overview

Agent	Best For	Allocated Skills (summary)
`@orchestrator`	Multi-phase coordination	Language skills + utility skills + `blogger`/`brutal-critic`
`@planner`	Read-only architecture/planning	Language skills + utility skills
`@codebase`	Feature implementation	Language skills + `sql-migrations`
`@review`	Security/performance/code quality	Language skills + `docs-validation` + `agent-diagnostics`
`@docs`	Documentation updates	`docs-validation` + `project-bootstrap` + `agent-diagnostics`
`@em-advisor`	Engineering leadership guidance	`project-bootstrap` + `docs-validation` + `agent-diagnostics`
`@blogger`	Blog/video/podcast drafts	`blogger` + `brutal-critic`
`@brutal-critic`	Final content quality gate	`brutal-critic` + `blogger`

See exact allowlists in the Skills Matrix.

Suggested Flow

@orchestrator (plan)
→ @codebase (implement)
→ @review (audit)
→ @docs (document)

Agent ↔ Skill Relationship (Quick Diagram)

Tip: use the +/−/⊙ controls to zoom, or scroll-wheel + drag to pan.

flowchart TD
    U["User Request in Claude"] --> O["orchestrator"]
    U --> P["planner"]
    U --> C["codebase"]
    U --> R["review"]
    U --> D["docs"]
    U --> E["em-advisor"]
    U --> B["blogger"]
    U --> X["brutal-critic"]

    subgraph Skills["Skills (.claude/skills)"]
      L1["Language Skills<br/>python/typescript/go/..."]
      L2["Utility Skills<br/>docs-validation/agent-diagnostics/project-bootstrap"]
      L3["Content Skills<br/>blogger/brutal-critic"]
      L4["Workflow Skills<br/>api-docs/code-review/..."]
    end

    O -.loads on demand.-> L1
    O -.loads on demand.-> L2
    O -.loads on demand.-> L4
    C -.loads on demand.-> L1
    C -.loads on demand.-> L4
    R -.loads on demand.-> L1
    R -.loads on demand.-> L2
    D -.loads on demand.-> L2
    B -.loads on demand.-> L3
    X -.loads on demand.-> L3
    P -.loads on demand.-> L1

    K["Dashed edge = loads skill on demand"]
    class K legend
    classDef legend fill:#1f2937,stroke:#4b5563,color:#e5e7eb

If Mermaid does not render in your docs host, the flow still reads top-to-bottom:

@orchestrator/@planner/@codebase/@review/@docs/@em-advisor/@blogger/@brutal-critic → load relevant skills on demand.

Skill Usage Guardrails

All built-in agents support skills.
Skills are loaded on demand (not eagerly).
Use one relevant skill per phase by default; add another only for clear cross-domain dependencies.
If stack/domain is unclear, clarify before loading.

Permission Model

Use .claude/settings.json for shared permission policy.
Use hooks for deterministic enforcement of risky operations.
Keep agent tool lists minimal and role-specific.

Skill Scope Policy (Current)

Keep current core-only skill scope.
Add skills only with repeat demand, clear gap, owner, and clean licensing/provenance.