polkit AI Reproducer Tools

Skills and environments for AI-powered polkit issue triage and reproduction.

Structure

• GEMINI.md — Agent instructions for Gemini CLI
• skills/ — Domain knowledge organized by component
• environments/ — Dockerfiles with systemd + polkit + Gemini CLI
• .github/workflows/build-images.yml — Builds and pushes container images

How it works

When a maintainer comments @ai-triage on a polkit issue:

1. Gemini REST API triages the issue (assess, label, ask for info)
2. A Docker container is started with systemd + polkit running
3. This repo is cloned into the container
4. Gemini CLI runs inside, iterating on a reproducer until it works
5. The reproducer is posted to the issue
6. If the agent learned something new, it opens a PR to add a skill file

Skills

Skills are markdown files containing domain expertise. They grow over time as the agent handles more issues. Each skill documents a technique for reproducing a specific class of polkit bug.

Container images

Pre-built images are available at:

• ghcr.io/polkit-org/polkit-ai-reproducer-tools:fedora
• ghcr.io/polkit-org/polkit-ai-reproducer-tools:ubuntu