RUOpen App
Documentation menu

Delegate issues to a coding agent and review its work

Updated October 22, 2026

You have a board of small, clear issues — a bug fix, a missing test, a settings page tweak — and no time to do them one by one. Hand them to a coding agent that runs on your own computer: it takes each issue, does the work on a branch, and tells you when it's ready for you to look at. You stay the reviewer.

This is different from chatting with an agent about your backlog: there you sit in the conversation; here you hand work off and come back to review it.

When it fits

  • Issues with a clear outcome you can check: "the export skips archived cards", "add a test for an empty list", "rename the setting in the UI and the docs".
  • Work in repositories you have cloned on your computer — the agent works there, with your build tools.
  • Not a fit: vague ideas, design decisions, anything you'd need to discuss first. Talk those through with the agent in a chat instead.

Set it up once

  1. Install the runner — one command, then click Allow in Boards. Set up the runner has the command for macOS, Linux and Windows.
  2. Link the repositories to the space. In the space's settings, add each repository's URL under Links, with a line on what it is ("the web app", "the API"). The runner finds your local clones by these links — nothing to map by hand. See Description and links.
  3. Choose where the agent works. If you review in your own editor, pick your own clones in setup: the agent's branch is then checked out right where you work. Where the agent works explains both options.
  4. Optionally, let the runner push the branch and open a pull request when the agent is done.

Write issues the agent can finish

The agent works only from what the issue says, so write it like a short brief:

  • What should be true when it's done — the behaviour, not the steps.
  • Where to look — the page, the endpoint, the file, if you know it.
  • How to check it — the test to add or the case to try.

If one issue has to wait for another, link them with blocked by — the agent doesn't take a blocked issue until its blockers are done. See Issue links.

Hand off and get on with your day

On each issue, click Hand to agent in the Agent section. The card shows For agent, then Agent working once your runner picks it up — and your computer shows a notification. While the agent works, the issue is locked so the task doesn't change under it; comments and status changes still work.

When it's done, the card shows Agent done · review and you get a ready for review notification. Clicking it opens the issue, with the agent's summary of what it changed.

Review where you work

The work is on a branch named after the issue, like agent/WEB-42-export-skips-archived-cards:

  • with your own clones, it's already checked out — open your editor, run the app and the tests;
  • with pull requests turned on, the link is in the runner's comment and the notification.

Ask for changes

Something isn't right? Ask the agent directly, in a terminal:

lba continue

It opens the agent's own session of the issue — the same conversation, on the same branch — so you can say "the button is still blue" and watch it fix it, without anything moving in Boards. Or hand over one change and let it work alone:

lba continue WEB-42 also hide the export button for read-only members

When you're done, one comment with the new commits goes to the issue. Away from your computer? Use Send back to agent on the issue instead: write what's wrong, and the agent picks it up again on the same branch. Ask for changes has the details.

Finish

Merge the branch or the pull request and move the issue to done. The card keeps a Done by agent badge, so you can always tell which work the agent did.

Keep it under control

  • Only your issues. The runner works only on issues assigned to you and handed off by you — nobody else can send work to your computer.
  • Nothing leaves your computer unless you say so. The agent can't push; the runner pushes only if you turned that on.
  • Your uncommitted changes are never touched. In your own clones, the runner asks before switching a clone that has uncommitted changes.
  • Stop any time. The × next to the agent badge cancels the run and unlocks the issue.
  • Everything is visible. Each run, and every change it made, is in the issue's history with the agent's API key.