Agents Panel

The Agents Panel lets you view and manage AI agents. Each agent is a specialized persona defined by a Markdown file with a system prompt and capabilities.

Built-in Agents

🔬 Analyst

Requirements analysis, research, and specifications.

🏗️ Architect

System design, API design, and technical decisions.

💻 Developer

Code implementation, debugging, and refactoring.

🔍 Reviewer

Code review, security audit, and best practices.

🧪 Tester

Test generation, verification, and coverage.

📊 Manager

Project management, task coordination, and estimation.

📚 Researcher

Deep research, knowledge gathering, and evaluation.

Custom Agents

Agent definitions are Markdown files in the workspace's agents/ directory:

analyst.md

architect.md

developer.md

frontend-reviewer.md

agents/frontend-reviewer.md

# Frontend Reviewer

You are a senior frontend developer specializing in React and TypeScript.

## Responsibilities

- Review React component architecture
- Check for accessibility issues
- Verify TypeScript type safety

## Rules

- Always check for proper error boundaries
- Verify hooks follow the Rules of Hooks

Use reload_agents to hot-reload agent definitions after editing files on disk.

Execution Modes

Full execution with file and bash access. Uses Claude Code CLI or AI SDK under the hood.

Spawn the architect agent to design a caching layer for our API.

LLM-only execution — fast but no file or bash access. Great for analysis and planning.

Run a quick agent (analyst) to compare REST vs GraphQL for our use case.

Run multiple agents concurrently. Each gets an equal share of the remaining budget.

Run reviewer and developer agents in parallel on the auth module.

Budget

Each agent run consumes from the daily LLM budget. Use agent_budget to check current usage.