agentleFS
h3y6eh3y6e/agent-skills/skills/orchestrating/SKILL.md

orchestrating

Skill0 starsChanged 28 days agoMITView on GitHub

Coordinates parallel coding work from a bare repository by running one agent session per worktree in Rex and tracking them. Use when started in a bare repository and asked for code changes, or asked about work already handed off from one.

4 other projects use a version of this.

Adopted by

Most recently taken up by

The file

Files
SKILL.md

Orchestrating

You are the coordinator: you never edit code. The tabs and git wt (no arguments) are the task list, not your memory.

Start a Child

  1. git wt <branch> origin/<base> --nocd.
  2. With rex run --label <branch>, open a tab in the worktree, without taking focus, and start the agent with the brief as its launch argument. A brief typed after launch can land in a trust or permission prompt. The label maps the branch to its block ID in rex block list.
  3. The child cannot see this conversation, so the brief stands alone, including the base branch. Ask it to follow the git-shipping skill through the PR and to end with a report: PR URL, evidence, and who acts next.
  4. Capture the tab once. A startup prompt such as folder trust reports no status, so the child looks idle while it waits. If one shows, ask the user to answer it in that tab.

Track

A child's state is its pane's program status: rex block call -b <block> program_status gives idle, working, blocked, done, or error. If you can watch a command in the background, watch this stream:

rex events -s "$REX_SESSION" --json | jq --unbuffered -c '.payload.event? | select(.type == "block_event" and .payload.name == "program_status_changed" and .payload.block_id != env.REX_BLOCK) | {block: .payload.block_id, state: .payload.payload.record.state} | select(.state | IN("done", "blocked", "error"))'

The stream reports only changes, so check every child once when you start it. Otherwise, check every child each time the user talks to you.

A status disappears when the agent exits and leaves a shell, so an idle child may be gone; confirm its agent is running before calling it waiting or sending to it.

Follow-ups

Send follow-ups only to an idle or done child. Never send to a blocked one: Enter on a permission prompt approves it. Its approvals and questions belong to the user; name the tab instead of answering.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.