I been using LLMs for coding a lot and while they aren't awful at generating code they suck at debugging and spend endless tokens on a debugging loop. Eventually that frustrated me and I built a CLI that attempts to make AI debugging easier. How it works is any error you encounter you would run a simple command "paper run <cmd>" which would "capture" that error and log it in a incident_report and tell your agents.md that is the source of truth for the error. When you ask your agent to fix the error the agent will look at your incident_report and solve it. Once your done you can shelf your report. Currently the constraints are it does not prompt a LLM or call external APIs(only writes a local .md file), everything stays local, no background daemon and no third party packages. Thanks for reading. The npm package page is posted below if anyone is interested. https://www.npmjs.com/package/@varman96/paper