Configure a coding agent

Use any coding agent that can run through a compatible local command. Codex CLI, Claude Code, and Cursor CLI are documented examples, not an exclusive list. To bring another agent, write a coding-agent command that passes the required {alterac-ai-message} placeholder as its initial instructions. The agent must be able to receive and follow those task instructions and use the alterac.ai CLI workflow to load and complete the exact work run. A command-line interface alone does not establish compatibility; verify your chosen program, version, and command.

Create a reusable local command that alterac.ai can offer when you queue work. The command runs on your computer with your operating-system user permissions. alterac.ai does not sandbox it, reduce its permissions, or run it for you.

This takes about 10 minutes. You need an installed and authenticated coding-agent program, an initialized alterac.ai repository, and permission to manage the definition at the scope you choose. For installation and sign-in, use your provider's instructions: Codex CLI (external site) or Claude Code (external site). Other compatible programs use their own documentation. These are provider-specific command examples, not native integrations or a guarantee that every coding-agent program is compatible. This is a prerequisite for completing your first task; it does not install or configure a coding-agent program for you.

Your alterac.ai license does not include coding-agent subscriptions or usage. Any costs from the program or provider you choose are separate, including when you use free individual access to alterac.ai.

1. Open Coding agents from your account menu to create a personal entry, or open Coding agents in an organization or project to create a shared entry.
2. Select Create coding agent and give it a recognizable name, such as Codex.
3. Enter your command in Command template, including the quoted {alterac-ai-message} placeholder.
   • You can write your own command or start with an example coding-agent command. Check the example's tested CLI version, selector choices, and permissions before using it.
4. Save the definition. The first enabled definition at a scope becomes that scope's default when there is no available default already.

The exact, lowercase placeholder is required for the launch contract. alterac.ai replaces it only when it launches a queued run, so the agent first receives repository-process routing, its labeled execution ID, and the exact command that loads immutable context. The command field's hint text explains this replacement boundary. If a saved command or queued snapshot lacks this placeholder, add it to the definition and requeue the task.

Personal entries belong to you. To make one available in a project, the organization must allow personal coding agents. For a command everyone in an organization can use, create an organization entry. For one project only, create a project entry when organization policy allows it. Choose where a coding agent belongs explains the available scopes and policy controls.

Queue a task and select the enabled definition. From the repository that is configured for that project, start the listener:

alterac-ai listen

With the recommended approve-changed trust mode, inspect the displayed command before approving it. The listener starts the program in a separate terminal when it can; that terminal owns the agent's output. If it cannot open a terminal, copy the printed command into one yourself.

Your task remains Queued while it waits for an approval. It changes to Agent working only after the listener has approval and starts the run. Run the local task listener explains these states and recovery options.

You can use CodeGraph alongside alterac.ai through your coding agent's existing MCP connection or its CLI. Configure the tool separately and give each alterac.ai task worktree its own index. Your coding-agent launch command stays the same. Use CodeGraph with a coding agent explains the worktree requirements and links to CodeGraph's installation and MCP setup documentation.

Choose a command trust mode before unattended use, then review command-template rules before adding options or another provider.