Launching agents

A step-by-step guide to launching and dispatching AI agents in ForgeDeck.

Embed
Preview of the card
<a href="https://docs.forgedeck.net/launching-agents"><img src="https://docs.forgedeck.net/_api/public/orgs/db9cb8ff-b606-46cf-bbcd-02c8f8a9a3e2/spaces/6f417400-93ca-41a8-a914-df2ae3981dc6/pages/b297b112-ced6-417b-b0a4-a43e1c3c0f85/social-card.svg?v=3e7ae33eec8b" alt="Launching agents" width="600" height="315"></a>

This guide walks the everyday loop: turn a task into a running agent and bring its work back to your main branch.

1. Describe the task

Add a card on the board. Give it a clear title and any notes the agent needs. Optionally set a file scope it's allowed to touch.

2. Pick an agent

Choose an agent profile (Claude Code, Codex, Gemini, …). ForgeDeck only offers CLIs it detects on your machine.

3. Launch

Drag to Running. ForgeDeck:

  • creates a worktree on a task branch,

  • opens a task-pinned terminal there,

  • writes the prompt bundle and sends the editable command.

Tweak the command if you like, then let it run. Output streams live in the pane.

4. Dispatch follow-ups

Already have an agent running in a pane? Dispatch a new task to it — ForgeDeck sends a one-line brief pointing at a fresh prompt file, keeping the same conversation (no new terminal). You can also resume a conversation into a different agent.

5. Review & merge

When the agent finishes, the task moves to In Review. See Review & merge.

Updated