Specialized sub-agents

Purpose-built agents with scoped context, tools, and termination conditions.

Atomic ships 12 specialized sub-agents alongside skills and workflows. Each is a purpose-built agent with a narrow job, a scoped tool set, and its own context window.

Skills and workflows auto-dispatch them. You can also invoke one directly with:

Task(subagent_type: "<name>", description: "...", prompt: "...")

Run /agents in any chat to list them.

The 12 sub-agents

Sub-agent	Purpose
planner	Decompose specs into structured task lists with dependency tracking.
worker	Implement single focused tasks (multiple workers run in parallel).
reviewer	Audit implementations against specs and best practices.
orchestrator	Coordinate complex multi-step workflows.
debugger	Debug errors, test failures, and unexpected behavior.
code-simplifier	Simplify and refine code for clarity and maintainability.
codebase-locator	Locate files, directories, and components.
codebase-analyzer	Analyze implementation details of specific components.
codebase-pattern-finder	Find similar implementations and usage examples.
codebase-online-researcher	Research using web sources and DeepWiki.
codebase-research-locator	Find prior research documents in research/.
codebase-research-analyzer	Deep-dive on existing research topics.

Why specialized agents instead of one general agent?

LLMs have an architectural limitation: the more context they hold, the harder it is to attend to the right information. A single agent juggling a spec, dozens of files, tool outputs, and its own reasoning will lose details, repeat work, or hallucinate connections.

Specialized sub-agents fix this with three constraints:

• Context isolation — fresh, minimal context per job. A worker doesn’t see the planner’s deliberation; a reviewer doesn’t see the worker’s scratch work.
• Tool scoping — a reviewer can’t edit files; a worker can’t spawn other workers. The available tool set matches the job.
• Parallel execution — independent agents run concurrently. Several codebase-locator runs can fan out without contaminating each other’s context.

The research-codebase skill, the create-spec skill, and the ralph workflow all rely on these sub-agents under the hood. You don’t usually invoke them by hand — they’re dispatched as your higher-level skill or workflow runs.