Claude Code subagent imported from oimiragieo/agent-studio (
.claude/agents/CLAUDE.md). Copyright stays with the author.
Agents
The agent system contains 119 specialized AI agents organized into 4 tiers. Each agent is a markdown file defining its role, tools, model preference, and behavioral instructions. Agents are spawned by the router via Task() — they never run independently.
Subdirectories
| Directory | Count | Purpose |
|---|---|---|
core/ |
10 | Essential agents used in every pipeline — router, planner, developer, QA, architect, etc. |
domain/ |
68 | Technology-specific specialists — one per language, framework, or domain (python-pro, kubernetes-specialist, etc.) |
orchestrators/ |
16 | Multi-agent coordinators — domain routers, swarm coordinator, master orchestrator, etc. |
specialized/ |
25 | Cross-cutting concern specialists — code-reviewer, security-architect, devops, researcher, etc. |
How Agents Work
- Definition: Each
.mdfile has YAML frontmatter (model,tools,description) followed by system prompt instructions. - Spawning: Router calls
Task({ subagent_type: 'agent-name', prompt: '...' }). - Registry: All agents are indexed in
.claude/context/agent-registry.json(source of truth for routing). - Iron Law: The
developeragent is the LAST RESORT — always use a specialist if one fits.
Related
.claude/context/agent-registry.json— Agent lookup registry.claude/docs/@AGENT_ROUTING_TABLE.md— Full routing matrix.claude/rules/agents.md— Quick-reference routing rules