Connect your coding agents

In the repository you want agents to work on, preview the changes, then apply them:

agentd init --dry-run
agentd connect --dry-run
agentd init
agentd connect
agentd connect --repo-rule   # optional
agentd doctor

Command	What it does
agentd init	Creates the local ledger and the systemd user services.
agentd connect	Registers the agentd MCP server and hooks in Claude Code and Codex, and backs up every file it changes. To connect one client, use agentd connect claude or agentd connect codex.
agentd connect --repo-rule	Adds an untracked rule to this repository telling your sessions to send implementation work through AgentD. It is optional and never edits a tracked file.
agentd doctor	Checks the setup and prints the command that fixes each failure. Run it until it reports no required failure.

Open a new Claude Code or Codex session afterwards. claude mcp list should show agentd as connected. For Claude workers, run claude in the repository once and trust it.

What a session gets

• The MCP server. Each session joins the shared ledger and starts with a short list of tools: its inbox, claiming, checkpointing and finishing a handoff, file claims, and approval requests. tools_more lists the rest.
• The hooks (Claude Code). They deliver the inbox and context reminders, warn on edits to claimed paths, and keep the session's checkpoint current.
• Two skills. agentd hands work to a worker from the session: it drafts a task from the code (goal, acceptance criteria, verification), shows it before starting it, checks progress and handles the Inbox. agentd-setup takes a session from an installed binary to a first verified task.

init and connect are safe to run again: a second run changes nothing.

Set up from a session

AgentD also works the way you already use Claude Code or Codex. In a session in your repository, ask:

Set up AgentD for this repository. Connect my coding agents and help me run my first verified development task.

The session runs agentd init, connect and doctor, and asks before it starts a worker, since a worker costs model usage.

Next: The dashboard.