Updated
Updated · InfoWorld · Sep 22
GitHub Issues Coordinate AI Agents Across 2 Machines and Projects
Updated
Updated · InfoWorld · Sep 22

GitHub Issues Coordinate AI Agents Across 2 Machines and Projects

3 articles · Updated · InfoWorld · Sep 22

Summary

  • GitHub issues can serve as a shared protocol for AI agents, letting Claude Code and Codex coordinate work across agents, machines and repositories while keeping exchanges visible and auditable.
  • Bram pairs local git and GitHub CLI access with indexed sessions, commits, worklists and issues, giving agents searchable context and a common place to review each other’s work or hand off tasks.
  • On separate Windows and Mac machines, the setup lets one agent file a bug with trace evidence and another pick it up; across projects, one Bram instance can open an issue in XMLUI and another can fix and close it.
  • A direct local API path let Codex relay Claude’s words under Claude’s name, prompting the author to require explicit agent signatures and keep coordination on the record rather than in hidden agent-to-agent chats.
  • The approach is framed as a practical alternative to formal JSON-based agent protocols such as A2A: slower by design, but easier for humans to inspect, steer and search.

Insights

Can legacy tools like GitHub issues truly cage autonomous AI, or are developers just creating a false sense of control?
When AI agents debug each other across machines, what stops them from bypassing strict identity rules entirely?
If AI agents secretly spin up headless models in your local repo, who is really writing your code?