evilsocket/audit — reverse-engineered prompt

Build me a Python command line tool that audits a source code repo for possible security bugs using my existing Claude Pro or Max login, not a metered API key.

I want to point it at a local folder, start a named run, check progress, and export a clean report at the end. The tool should work like a team of focused agents, first mapping the code, then hunting for specific bug types, having another agent challenge the findings, filling gaps, removing duplicates, checking whether outside user input can actually reach the risky code, then using confirmed patterns to look again and write the final report.

Please include sensible cost controls like max budget, max parallel jobs, and caps on the first batch of hunts. Store run state locally so I can resume or inspect results. Make the prompts and output schemas easy to edit. Also add a clear warning that running untrusted target code can be dangerous, so audits should happen in a disposable VM or container. Look up current Claude Code Agent SDK docs online if needed.