This guide walks you through spawning your first Claude Code agent in Pentagon.
Spawning an Agent
- Click any empty grid cell on the Office Canvas, or press Cmd + N if a spawn dialog shortcut is bound.
- Select a project directory — the folder Claude will work in.
- Name your agent — Pentagon auto-generates cosmic names (like “nova-pulse”), or enter your own.
- Choose a model:
- Opus — most capable (default)
- Sonnet — balanced speed and capability
- Haiku — fastest responses
- Set permission mode:
- Full Auto — Claude runs tools without asking (bypass mode)
- Custom — select which tools are allowed (Read, Edit, Bash, Write, Grep, Glob)
- Choose branch strategy:
- Use existing branch — work on your current branch (default for the first agent in a repo)
- Create new worktree — isolated branch for this agent (default for subsequent agents)
- Click “Create”
Your agent appears as a desk on the canvas and begins its Claude Code session.
Interacting via Terminal
To interact with an agent:
- Click the agent’s desk on the canvas to select it
- The Detail Panel opens on the right
- Switch to the Terminal tab to see the live Claude Code session
- Type your prompt and press Enter to send
What files are in this directory?
Press Cmd + L to focus the terminal input for the selected agent.
Agent Status
Each agent desk displays a colored ring indicating its current state:
| Status | Ring Color | Description |
|---|
| Ready | Green | Idle — waiting for your input |
| Active | Yellow | Claude is working (tool calls in progress) |
| Needs Input | Green | Finished its turn — ready for your next prompt |
| Dormant | Gray | Paused — session preserved but not running |
| Error | Red | Something went wrong |
Stopping and Resuming
- Terminate: Select the agent desk and press Delete (or Backspace). Pentagon asks for confirmation, then kills the process and removes the agent.
- Respawn: Use the agent’s context menu to respawn — this kills the current process and starts a fresh session. The agent’s soul, tasks, and memory persist.
- App restart: When you quit and reopen Pentagon, all agents restore to their previous state. Sessions resume where they left off.
Cloning an Agent
Adjacent to each agent desk, you’ll see + indicators on empty grid cells. Click one to clone that agent — Pentagon creates a new agent with the same project directory, automatically isolated in its own git worktree.
Tips
- Be specific — clear, detailed prompts get better results
- Use worktrees — let Pentagon isolate each agent’s changes automatically
- Review changes — always review code changes before merging
- Watch the rings — status colors give you instant feedback across all agents
Next Steps
