quadrant: reference audience: agents, maintainers see-also: - orchestrator-usage.md - ../references/agent-routing.md - ../references/agent-authoring.md - cli-reference.md

Agent roster¶

Pillar context

This page details Pillar 2 — Compose the right team. For the high-level map of all four pillars, see The Four Pillars.

This page mirrors agents/*.md — the distributable agent definitions installed by scripts/install.sh. There are 30 agents. The orchestrator picks among them based on task domain, risk tier, and budget. To dispatch one directly inside Claude Code, name it in the Agent tool with subagent_type.

To inspect runtime registration: baton agents.

For routing rules: references/agent-routing.md.

How agents are invoked¶

Agents do not run on their own. They run when:

The orchestrator agent dispatches them as part of a baton-driven plan, OR
A user (or another agent) names them via Claude Code's Agent tool with a subagent_type parameter.

Agent files in agents/ use a YAML frontmatter block and a body prompt. The frontmatter is what the runtime registers; the body is what the agent reads when dispatched. New or updated/generated agents should include name, description, model, permissionMode, and tools, and should follow the generated-agent contract in references/agent-authoring.md.

Orchestration & routing¶

Agent	Model	Use when
`orchestrator`	opus	Complex cross-cutting tasks, multi-domain refactors, batches of related tasks. Drives `baton plan` → `baton execute`.
`team-lead`	sonnet	Coordinator for parallel sub-teams within a phase.
`task-runner`	haiku	Procedural execution of pre-scripted shell/HTTP/data-format tasks. No architectural judgment.

Implementation specialists¶

Agent	Model	Use when
`backend-engineer`	sonnet	Server-side: APIs, business logic, ORM, middleware, server config.
`backend-engineer--node`	sonnet	Node.js/TypeScript backend specifics (Express, Fastify, NestJS, Prisma, Drizzle).
`backend-engineer--python`	sonnet	Python backend specifics (FastAPI, Django, Flask, SQLAlchemy, Alembic, Pydantic, Poetry/uv).
`frontend-engineer`	sonnet	Client-side: components, styling, state, routing, forms, accessibility.
`frontend-engineer--react`	sonnet	React/Next.js: Server Components, Suspense, App Router, Zustand/Redux Toolkit/Jotai.
`frontend-engineer--dotnet`	sonnet	Blazor, Razor Pages, ASP.NET MVC views; SignalR.
`data-engineer`	sonnet	Schemas, migrations, query optimization, ETL, data modeling.
`devops-engineer`	sonnet	CI/CD, Docker, deployments, environment config, build optimization.
`architect`	opus	Data-model design, API contracts, technology selection, module boundaries, second-opinion reviews.

Review & quality¶

Agent	Model	Use when
`code-reviewer`	sonnet	Final pass for readability, performance, error handling, project conventions.
`security-reviewer`	opus	Auth flows, input validation, secrets management, OWASP top 10, dependency vulns.
`auditor`	opus	Independent safety/compliance/governance review. Has veto authority. Required for MEDIUM+ risk plans.
`test-engineer`	sonnet	Unit, integration, and E2E tests. Test infrastructure.

Data & analysis¶

Agent	Model	Use when
`data-analyst`	sonnet	Business intelligence, reporting, SQL queries, KPI definition, dashboard design.
`data-scientist`	sonnet	Statistical analysis, ML modeling, experiment design, model evaluation.
`visualization-expert`	sonnet	Chart design, dashboard layout, visual storytelling.

Team composition¶

Agent	Model	Use when
`talent-builder`	opus	Pillar 2 — compose the right team. Researches a domain and creates a bespoke specialist agent + knowledge pack + skills, so no generalist drowns in whole-codebase context.

Domain & governance¶

Agent	Model	Use when
`subject-matter-expert`	opus	Industry-specific business rules, regulatory compliance (SOX, GDPR, HIPAA), business processes.
`learning-analyst`	sonnet	Reads execution history + scorecards; proposes evidence-backed agent/config improvements.
`system-maintainer`	sonnet	Post-cycle config tuning. Mutates `learned-overrides.json` only — never source code.

Archetypes¶

Agent	Model	Use when
`archetype-james-engineering-manager`	opus	Evaluating Agent Baton features from a management perspective: PMO dashboard UX, governance/approval workflows, cost visibility, reporting, and analytics. Use to answer "would an engineering manager actually use this to sponsor and oversee an agent program?"

Resilience subsystem (immune)¶

Agent	Model	Use when
`immune-autofix`	haiku	Applies high-confidence hygiene fixes found by the immune sweep (stale comments, formatting, trivial lint). Invoked automatically by `FindingTriage` when `auto_fix=True`.
`immune-deprecated-api`	sonnet	Sweeps for usage of deprecated APIs flagged by upstream.
`immune-doc-drift`	sonnet	Detects when docs disagree with source.
`immune-stale-comment`	haiku	Finds stale comments and TODOs that no longer match the code.
`immune-todo-rot`	haiku	TODO/FIXME age detection and triage.
`immune-untested-edges`	sonnet	Identifies code paths missing test coverage.

For the routing logic that picks among these agents, see references/agent-routing.md. For the generated-agent authoring contract, see references/agent-authoring.md. For risk-tier guardrails, see references/guardrail-presets.md. For each agent's full prompt, read the matching file in agents/<name>.md.