GitHub Code Monkey Bot

Scheduled AI authoring of pull requests for open GitHub issues. A reusable workflow selects the highest-ranked triaged issues, runs one Copilot CLI coding agent per repository against a checkout of that repository, and publishes the result as GitHub-signed commits on a branch in a bot fork organisation, with a pull request against the target. Nothing merges without a maintainer.

How a run works

select (trusted)
  | ranked issues, one per repository, base commit, guidance
  | commit SHA, evidence ID, digests
  v
author (untrusted, one runner per issue)
  | checkout + Copilot CLI; no credential that can write
  | git bundle + manifest
  v
publish (trusted, one runner per issue)
  | verify the bundle offline
  | fork token: find or create the fork, push the branch
  | target token: open the pull request, comment on the issue
  v
report (trusted)

Where bot branches live

The publisher never pushes a branch to a target repository. It pushes to a fork held in a bot fork organisation and opens the pull request from there, so the target's workflows run the change as an outside contribution, with no secrets and an approval gate.

Target organisation	Fork organisation
onap	lfreleng-bot-forks-onap
opendaylight	lfreleng-bot-forks-opendaylight
o-ran-sc	lfreleng-bot-forks-oransc
every other organisation	lfreleng-bot-forks

Where to go next

• Design: the architecture, trust model, selection, signing, publishing, credentials, rollout and data contracts.
• Guidance for agents: the rule every AI agent in the organisation follows when it opens a pull request.