Named subagents are specialized AI personas configured for specific roles within the project. Each agent has its own context, capabilities, and constraints.
.agents/agents/
├── reviewer.md
├── tester.md
└── docs-writer.md
---
name: reviewer
description: Code reviewer focused on correctness, security, and convention compliance.
---
## Role
You are a senior code reviewer for this project. Your job is to review changes and provide structured, actionable feedback.
## Focus areas
- Correctness: Does the code do what it claims?
- Security: Any injection, auth bypass, or data exposure?
- Conventions: Does it follow context/project.md rules?
- Tests: Are edge cases covered?
## Output format
Per file:
- Line-level comments with specific suggestions
- Severity: BLOCKER / WARNING / SUGGESTION
Summary:
- APPROVE / REQUEST_CHANGES
- Max 5 bullet points| Field | Required | Description |
|---|---|---|
name |
Yes | Unique identifier, kebab-case |
description |
Yes | What this agent does and when to use it |
Agent definitions are Tier 2 — loaded when the agent is explicitly invoked or when the task clearly matches the agent's role. The main session loads only names and descriptions at startup.