krishkuchroo/orion ? reverse-engineered prompt

Build me a standalone code security scanner that I can point at any Git repo and have it find real vulnerabilities with low false positives.

It should first build a code graph from the project, then use multiple Claude agent runs to look for security issues in different ways, and then use a separate verifier agent to independently check each lead before reporting it. I want the tool to work across different stacks, with automatic language detection when it can, and a generic fallback when it can’t recognize the framework.

Please include a simple command line flow where I can run one scan on a local repository, watch live progress while it runs, and get a ranked report at the end with evidence for each finding. If it helps, look up current docs online for Joern, Claude Code, and Neo4j. Also make sure the scan can be saved and resumed, and that results can be output as JSON too.