codescan

Finds the code AI pretended to write.

Multi-language code graph plus AI-slop detection: tree-sitter parses the tree, a duplicate-safe graph is built, and the report scores what it finds — unfinished-work markers, narrative stand-ins for implementations, mock data on live paths, duplicate definitions, uncalled functions. --dot-out renders the call graph. The report shape matches audit-web, so the scoring and PDF pipeline apply unchanged.