Inspiration
We are developers who have used LLM-based workflows for debugging in hacky ways. We wanted to smooth this process through automation and CLI loveliness.
What it does
Analyzes your Python script and runs it Captures stderr if the script exits with error Infers your code's purpose through conversational prompts Asks clarifying questions if initial purpose guess is wrong Determines if it can suggest a fix with available info Provides context-aware fix suggestions Allows the user to see an explanation of why the fix works.
How we built it
Claude + Langchain + Python.
Challenges we ran into
Differences between langchain wrapped claude calls compared to claude workbench wasting several hours. Merge confs Silly syntax errors.
Accomplishments that we're proud of
Made friends, made it look sexy, got a fairly functional output we will use outside of this weekend.
What we learned
Restricting output on claude using XML, langchain
What's next for FSD - Friendly Script Doctor
Lots of potential additions
- handling 0 error codes and providing refactoring suggestions.
- handling larger codebases
- prioritising context to send in to get better outputs
- re-running fixed code programmatically.
Built With
- claude
- langchain
- python
Log in or sign up for Devpost to join the conversation.