Agents

Lean reference for the built-in agents.

Why these agents

• Clear separation of roles so you can pick the right depth quickly.
• Safer behavior by default through scoped skill allowlists.
• 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)

Skill Usage Guardrails

• All built-in agents support the skill tool.
• 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.

Skill Permission Model

• Agents use permission.skill rules to restrict which skills can load.
• Recommended baseline is "*": "deny" with explicit per-skill allows.
• This enforces least privilege and keeps agent behavior domain-scoped.

Skill Scope Policy (Current)

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

Next Steps

• Commands & Skills
• Coding Standards
• Customization