Skip to content

DEBUGGER_AGENT.md

Latest commit

 

History

History
53 lines (38 loc) · 1.81 KB

DEBUGGER_AGENT.md

File metadata and controls

53 lines (38 loc) · 1.81 KB

Debugger Agent 🤖

An AI Agent that automatically resolves failing tests in CI

Overview

This is an agent that automatically debugs failing tests in CI. The fix for the failing test is added as a suggestion on a failing PR that can be accepted with a single click.

Check out the full demo below: Watch the video

Implementation

The agent is a Dagger function that automatically debugs failing tests in CI.

In the dagger module under .dagger directory, there is a new function called debug-tests.

The debug-tests function:

  • Creates a Workspace for an LLM to read and write the files in the project and run tests
  • Passes in the appropriate source and checker function to the workspace
  • Give the LLM a prompt to debug the broken tests
  • Get back a unified diff of the test fixes

To get a useful agentic flow out of this function, there's another function called debug-broken-tests-pr that:

  • Uses the GitHub API to get the PR number and the branch name
  • Uses the debug-tests function to debug the broken tests
  • Uses the GitHub API to comment on the PR with suggestions of the fixes

How do I try it?

The only dependency to run this agent is Dagger. Here are the installation instructions.

$ Fork or clone this repository and checkout the tricky_broken_tests branch:

git clone https://github.com/kpenfound/greetings-api
cd greetings-api
git checkout tricky_broken_tests

$ Get in a Dagger shell:

dagger

⋈ Run test function to see the failed tests:

test

⋈ Run debug-tests to let the agent fix the tests and tell you the fix.

debug-tests --model <any model, e.g. "gpt-4o">