Skip to content

Latest commit

 

History

History
51 lines (45 loc) · 2.08 KB

File metadata and controls

51 lines (45 loc) · 2.08 KB

explainer

how to ingest repos

questions

  • what type of errors is this best for?
    • syntax errors
    • logical errors
    • dependencies
  • how does an error explainer fit in the workflow of a typical user?
  • what is the minimum info you need to pass to the llm to explain the errors?
    • only logs
    • logs and source code
    • entry point
  • how do we deal with large code bases / logs
    • ran into token limits with the OpenAI API
  • what can we do with the output of the error explainer?
    • can we create a PR to fix the error?
    • can the explanation reference the exact line of code that caused the error?
  • cost
    • should the error explainer be available for:
      • paid accounts only?
      • every type of error?
    • how can we put a cost limit after which the explainer is not presented?
  • can we explain things other than errors by tweaking the prompt and inputs?
    • how the code works
    • code quality assessment
    • generate documentation
    • generate tests
  • how to run the error explainer in prod
    • repomix supports running as an MCP server

experiments

  • use some of the latest publish failures from production (public repos)
  • run them on my own account in prod to retrieve the logs
  • use repomix to get the source code
  • feed into the error explainer, along with the entry point
  • use this prompt
  • some early results