cloudflare/security-audit-skill ? reverse-engineered prompt

Build me a coding agent skill that can do a real security audit on a codebase in multiple passes, not just a quick scan.

I want it to first map the app, trust boundaries, and input paths, then have separate agents hunt for different kinds of bugs like injection, access control, business logic flaws, auth problems, client side issues, and anything else suspicious. After that, it should challenge its own findings and reject weak ones, then produce a clear human readable report plus a structured JSON file of the results. The JSON should be easy to validate against a schema, and the tool should verify each finding against the actual code before it counts as confirmed.

If there are previous audit results in the output folder, use them so later runs explore new areas instead of repeating the same work. Make it feel like a workflow I can trigger with something like security audit this codebase, and have it ask for an output folder if I do not provide one.