visa/visa-vulnerability-agentic-harness ? reverse-engineered prompt

Build me a Python command line tool that can scan a code repository for security issues, explain what it finds in plain language, and then suggest and validate fixes. I want it to work in stages, starting with understanding the code and threat model, then looking deeper for likely vulnerabilities, then deduplicating and reporting the best findings, and finally proposing minimal remediation changes and checking that the fixes actually hold up.

The output should include a human readable report and a SARIF file so the findings can be used in other security tools. It should support running in detection only mode without changing code, and a default mode that can go on to remediation and validation. Make it work with modern LLM providers through a flexible layer, and keep the whole thing usable from the terminal with helpful commands like setup, doctor, estimate, scan, remediate, and validate. If you need to, look up current docs online for the model or API integration details.