Bugpatrol is a QA team made of agents. It uses your web, desktop, or mobile app, HTTP API, or CLI the way a tester does. It finds bugs, fixes them, checks each fix in the running app, and opens the pull requests and issues for your team. The Nebula team uses it every day on our own apps, and we made it open source for all teams.
- An explorer agent runs the app, maps its screens, and reports what looks wrong.
- A judge agent decides which reports are real bugs, and writes the issues.
- A fixer agent writes a fix in its own git worktree.
- The explorer and the judge retest each fix in the running app.
- Bugpatrol publishes to GitHub: a PR for each fix, and an issue for each major bug with no fix.
npx bugpatrol
npx bugpatrol runs the patrol: explore, judge, fix, retest, and publish, again and again. The fixer and GitHub stay off until you turn them on.
flowchart TB
start(["npx bugpatrol"]) --> pull["Pull main and start the app"]
pull --> explorer{{"1 · Explore<br/>LLM · Explorer uses the app"}}
explorer -->|"bug reports"| judge
judge["2 · Judge<br/>LLM · Judge files the issues"] --> fixer["3 · Fix<br/>LLM · Fixer writes a fix"]
fixer --> retest["4 · Retest<br/>LLM · Explorer + Judge check the fix"]
retest --> publish["5 · Publish<br/>LLM · Judge opens PRs and issues"]
publish --> wait["Sync GitHub, then wait for a new commit on main"]
wait -->|"next cycle"| pull
subgraph legend ["Legend"]
direction LR
l1["LLM agent"] ~~~ l3["Code, no model"] ~~~ l4["Off by default"]
end
wait ~~~ legend
classDef llm fill:#312e81,stroke:#a5b4fc,color:#ffffff
classDef code fill:#1e293b,stroke:#64748b,color:#e2e8f0
classDef opt fill:#312e81,stroke:#a5b4fc,color:#ffffff,stroke-width:2px,stroke-dasharray:6 4
class explorer,judge,l1 llm
class start,pull,wait,l3 code
class fixer,retest,publish,l4 opt
style legend fill:transparent,stroke:#64748b
How it works tells when the patrol runs, and shows each step of the explore loop.
Set up with your agent
-
Install the Bugpatrol skill in your repo, and commit the skill and skills-lock.json:
npx skills add agent-labs-dev/bugpatrol
-
Give this prompt to your agent:
Set up Bugpatrol for this repo.
If your agent cannot install skills, give it the skill URL:
Read https://raw.githubusercontent.com/agent-labs-dev/bugpatrol/main/skills/bugpatrol/SKILL.md and set up Bugpatrol for this repo.
To set up Bugpatrol by hand, read Getting started.
What is supported
| Area | Supported |
|---|
| Platforms | Web (Playwright), Electron (CDP), iOS simulator and Android emulator (Maestro), Linux desktop apps through Cua and Xvfb, HTTP APIs and CLIs |
| Agent LLMs | Claude Code, Codex, Kimi CLI, opencode, pi, or any CLI agent; or an API key for OpenRouter, Vercel AI Gateway, OpenAI, Anthropic, or a custom endpoint |
| GitHub | PRs, issues, state sync, and pull request reviews in CI, through the gh CLI |
| Output | A local dashboard, GitHub PRs and issues, and JSON files under .bugpatrol/runs/ |
Documentation
| Page | What it covers |
|---|
| Getting started | The requirements, and the setup by hand, step by step |
| Commands | All the CLI commands |
| Configuration | All the settings in .bugpatrol/bugpatrol.yml |
| LLMs | The providers for each agent, and the model that we recommend for each agent |
| The dashboard | What each page shows, and the files behind it |
| How it works | The patrol, the cycle, noise control, memory, and safety |
| GitHub | PRs and issues, the state sync, and pull request reviews in CI |
| The deterministic gate | bugpatrol run: a merge gate for web apps that uses no agent |
| Development | Build Bugpatrol from source, the packages, and the examples |
| Bughunters | The earlier name of Bugpatrol, and how to move a repo to the new names |
License
Apache-2.0
