vercel-labs/deepsec ? reverse-engineered prompt

Build me a tool called deepsec that I can run inside a codebase to find security issues with the help of coding agents.

I want it to start by setting itself up in a repository, create a local working folder for scan data, then let me run a full scan, process the results with an AI agent, recheck findings, and export the final reports in a readable format. It should be able to handle big repos by splitting work up and resuming if a run gets interrupted, so I can just run the same command again and it picks up where it left off.

Please make it work well from the command line, with clear setup instructions, sensible defaults, and support for using my existing AI login or an API key if needed. If there are common frameworks or code patterns it should understand, include those too. Look up current docs online if you need to, and make the whole thing feel polished enough to use on a real production codebase.