Agents Reference

Agents are subagent definitions for Cursor and Claude Code. Each runs in an isolated context window.

Format

---
name: archon-example
description: When to use this agent.
model: inherit              # optional: fast, inherit, or specific model
readonly: true              # optional: restricts write permissions
skills:                     # optional (Claude Code): preload skills
  - archon-code-quality
---

System prompt / instructions in Markdown.

All Agents

archon-init

Field	Value
File	agents/archon-init.md
Trigger	/archon-init
Purpose	Bootstrap ecosystem or health check

archon-demand

Field	Value
File	agents/archon-demand.md
Trigger	/archon-demand <requirement>
Purpose	Full 7-stage delivery pipeline
Preloads	archon-code-quality, archon-test-sync, archon-async-loading, archon-error-handling

archon-audit

Field	Value
File	agents/archon-audit.md
Trigger	/archon-audit
Purpose	Project health check (0-100 score)
Mode	Read-only

archon-refactor

Field	Value
File	agents/archon-refactor.md
Trigger	/archon-refactor
Purpose	Progressive refactoring plan

archon-self-auditor

Field	Value
File	agents/archon-self-auditor.md
Called by	archon-demand Stage 3
Purpose	6-dimension code audit
Mode	Read-only
Preloads	All 4 constraint skills

archon-test-runner

Field	Value
File	agents/archon-test-runner.md
Called by	archon-demand Stage 3.4
Purpose	Test sync and execution
Preloads	archon-test-sync

archon-verifier

Field	Value
File	agents/archon-verifier.md
Trigger	/archon-verifier
Purpose	Independent work validation
Mode	Read-only

Agent vs Skill

Capability	Agent	Skill
Isolated context	✅	❌
Preload constraints	✅ (Claude Code)	❌
Read-only mode	✅	❌
Model selection	✅	❌
27+ tool support	❌	✅

Use agents when available. Fall back to skills for tools that don't support subagents.