Overview

The problem

With LLMs and fast iteration:

• Bugs discussed in chat are forgotten
• Tasks stay "mental" but never appear in the team backlog
• Scattered markdown files (TODO.md, bugs.txt) multiply
• Nobody knows what is open, in progress, or closed

The solution

Issuebeam connects AI agent chat to GitHub Issues with:

1. scripts/github_issue.py — stdlib Python CLI
2. Agent instructions — files the LLM reads (AGENTS.md, Cursor rule, Copilot, Claude, …)
3. GitHub templates — web forms for browser-based issue creation
4. adopt.py — copy the skeleton into any repo in one command

GitHub Issues becomes the source of truth. Local markdown stays for detailed plans and archives — not operational status.

Architecture

You in chat  →  AI agent (reads AGENTS.md)  →  github_issue.py  →  GitHub Issues

Requirements for any platform:

• Python 3 (Windows, macOS, Linux)
• GITHUB_TOKEN with Issues read/write
• Repo slug in tracker/github_repo or GITHUB_REPO
• Agent instructions: run the script, don't delegate to the user

Next steps

1. GitHub token
2. Repository slug
3. Adopt into a project (optional)
4. Choose your platform