Bright plans, routes, and verifies. Coding agents write the code. The handoff works when the issue can be executed without a round trip back to you, and it fails most often when the issue has no acceptance criteria. Open Bright →

​
What a good handoff contains

Before handing work to any coding agent, make sure the issue carries:
  • The problem and the desired user outcome
  • The repository and where the result should land
  • Existing behavior, with evidence
  • Constraints and explicit non-goals
  • Acceptance criteria, the loop’s exit condition
  • Security, migration, and rollout concerns
A handoff without acceptance criteria is the most common cause of a coding agent returning something plausible and wrong. It had nothing to verify against.

​
Hand an issue to GitHub Copilot’s coding agent

  1. Connect GitHub under Capabilities → Apps (see Connect apps).
  2. In a session, use /dispatch-coding-agent with a Bright issue.
  3. Bright dispatches the issue to GitHub Copilot’s coding agent, records the pull request link on the issue, and monitors the run.
When the pull request is ready, review it against the issue before it merges.

​
Drive Cursor

  1. Connect Cursor under Capabilities → Apps. Cursor uses a name and an API key from the Cursor dashboard rather than OAuth.
  2. In a session, mention Cursor or type /cursor to dispatch and monitor Cursor Cloud agents, runs, artifacts, models, and repositories.

​
After the handoff

Treat the coding agent’s output like any other result: check it against the acceptance criteria before you adopt it. Loop engineering covers the build-and-verify rhythm, Forge reference covers what its Acceptance phase checks, and Verify connector claims covers checking what a connector actually did at the provider.

​
Good to know

  • Both GitHub and Cursor connect under Capabilities → Apps.
  • Cursor connects with a name and an API key from the Cursor dashboard, not OAuth.
  • /dispatch-coding-agent records the pull request link on the issue and monitors the run.
  • Acceptance criteria are the loop’s exit condition. Without them the coding agent has nothing to verify against.

​
Next step

See where Bright executes code in Where your code runs.