Skip to main content
Full agent implementations that go beyond cookbook recipes. Each example is a complete project you can clone, configure, and run. All examples are available on GitHub.

​
Coding agent for GitHub issues

A multi-agent system that automatically analyzes GitHub issues and generates code fixes. Two specialized agents - a planner and a coder - share a Docker sandbox where they can read files, write code, run commands, and execute tests. What it does:
  1. Receives a GitHub issue (via CLI or webhook)
  2. Planner agent explores the codebase using sandbox tools (exec, read, glob, grep) and produces a fix plan
  3. Coder agent implements the fix using write and edit, then commits and pushes inside the Docker container
  4. Workflow suspends for human approval - shows the diff, commit log, and files changed
  5. On approval, creates a pull request and comments on the original issue
Polos features used:
  • Sandboxed execution - Both agents share a session-scoped Docker container with git credentials, network access, and built-in tools
  • Human-in-the-loop - Form-based approval with diff preview, rejection with feedback, 24-hour timeout
  • Durable workflows - 10 checkpointed steps. If the process crashes after the coder finishes, it resumes from the approval step - no agents re-run, no LLM calls repeated
  • Triggers - Can be triggered automatically via github/issues/opened events
Run it: 
git clone https://github.com/polos-dev/polos-examples.git
cd polos-examples/github-issue-fixer

cd python
cp .env.example .env
# Add your ANTHROPIC_API_KEY and GITHUB_TOKEN
polos dev
Then trigger it on an issue: 
python main.py owner/repo#123
Python source on GitHub | TypeScript source on GitHub