Agents & profiles

How ForgeDeck models AI coding agents — profiles, command templates, launch.

Embed
Preview of the card
<a href="https://docs.forgedeck.net/agents"><img src="https://docs.forgedeck.net/_api/public/orgs/db9cb8ff-b606-46cf-bbcd-02c8f8a9a3e2/spaces/6f417400-93ca-41a8-a914-df2ae3981dc6/pages/9d873914-d154-4fd6-991b-b8f94f041fc3/social-card.svg?v=3e7ae33eec8b" alt="Agents &amp; profiles" width="600" height="315"></a>

ForgeDeck is agent-agnostic: it drives whatever CLI you already use. An agent profile is the recipe for running one.

Profiles

A profile pins:

  • The command template — e.g. claude, npx codex, gemini, aider, or any shell command. Templates render variables (the task, the prompt file, the worktree path) before they're sent.

  • Defaults — model, flags, and how results are parsed.

ForgeDeck ships sensible defaults for Claude Code, OpenAI Codex, and Gemini CLI, and detects which are installed on your machine.

The launch flow

When you launch an agent on a task, ForgeDeck:

  1. Builds a context bundle and writes .forgedeck/runs/<task-id>/prompt.md.

  2. Opens a task-pinned terminal in the task's worktree (or project root).

  3. Renders the command template and sends the editable command — you can tweak it before it runs.

  4. Moves the task through Running → In Review/Needs Input → Merged/Done.

Bring your own keys

Agents run locally with your CLIs and API keys. Nothing about your code or keys leaves your machine for the core flow — see Security & privacy.

Updated