Inspiration

A dependency upgrade can pass existing tests and still break untested behavior. We wanted to help developers uncover those failures and remember verified findings.

What it does

Secondlook scans public GitHub repositories for supported compatibility risks and offers two prepared investigations showing failures and fixes across dependency versions.

How we built it

We built a Python application with a web dashboard. The live investigation connects Bright Data documentation retrieval, Strands/Groq test-input generation, Docker comparisons, and Cognee memory.

Challenges we ran into

Separating plausible AI suggestions from measured failures, handling service interruptions, and verifying that stored evidence could be retrieved accurately.

Accomplishments that we're proud of

We reproduced a pandas incompatibility in real repository code, exposed a Pydantic issue missed by existing tests, and verified fixes across both tested versions.

What we learned

Documentation guides investigation, tests establish behavior, and memory preserves evidence. Clear scope makes results more useful.

What's next for Secondlook

Expand supported upgrade patterns, generate tests beyond the prepared examples, and integrate investigations into pull requests and CI.

Built With

Share this project:

Updates