Claude Code subagent imported from TheLobbi/claude (.claude/agents/README.md). Copyright stays with the author.

Agent System

The agent system provides 137 specialized subagents that Claude invokes using the Task tool. Each agent is a markdown file with a YAML frontmatter block (or embedded YAML section) that declares its name, model, tools, and purpose. Agents are organized into domain-specific directories so they are easy to find and extend.

Why agents instead of one general assistant

A single general-purpose assistant accumulates context and loses precision on specialized tasks. Agents isolate concerns: a k8s-debugger knows Kubernetes deeply and nothing else, which keeps its instructions tight and its output reliable. The master orchestrators (angelos-symbo, ultrathink) handle decomposition and delegation — specialized agents handle execution.

Agent file format

---
name: agent-name
description: One-line description. Tells Claude when to invoke this agent.
tools: Read, Grep, Glob, Bash, Write, Edit
model: sonnet
memory: project
---

You are a [role]. [System prompt defining behavior, checklist, output format...]

Some agents use an embedded YAML block instead of frontmatter:

# Agent Title

## Agent Metadata
\`\`\`yaml
name: agent-name
model: opus
category: orchestration
keywords:
  - keyword1
  - keyword2
\`\`\`

The description field (or the keywords list) is what Claude reads to decide whether to invoke this agent. Write descriptions as "use this when..." phrases.

Top-level core agents (8)

These eight agents live directly in agents/ and are the most broadly applicable. They are proactively invoked across all domains.

Subdirectory catalog (28 categories, 129 agents)

Key agents spotlight

orchestration/ultrathink (model: opus) — Extended reasoning agent for problems that need deep deliberation. Invoked when you use keywords like ultrathink, think harder, or analyze deeply. Uses Claude's extended thinking API with configurable token budgets. Ideal for architecture decisions, security threat modeling, and multi-system analysis.

orchestration/angelos-symbo (model: opus) — Master orchestrator. Designs and coordinates complex multi-agent workflows. Handles task decomposition, parallel execution planning, context flow between agents, and graceful degradation when sub-agents fail. The recommended starting point for any task requiring 3+ agents.

orchestration/feature-dev (model: sonnet) — Full-stack TDD agent. Implements features end-to-end following the EXPLORE → PLAN → CODE → TEST → FIX → DOCUMENT protocol. Writes tests before implementation and does not consider a feature complete until all tests pass and documentation is updated.

Adding a new agent

1. Create agents/<category>/<name>.md
2. Add YAML frontmatter with name, description, tools, and model
3. Write a focused system prompt — one responsibility, clear output format
4. Keep the description field precise: it controls when Claude chooses this agent

Do not duplicate responsibilities across agents. If an agent would mostly repeat an existing one, extend the existing agent's system prompt instead.

See also

• ../README.md — Platform overview and component relationships
• ../skills/README.md — Skills that activate agents via triggers
• ../rules/README.md — Rules that constrain agent behavior