The Agent Skills Directory

[SAFE]: The skill operates entirely locally, processing text from output/CLAIMS.md and output/PAPER.md to generate analysis reports. The logic in scripts/run.py uses predefined regular expression patterns to search for specific terms (e.g., "ablation", "benchmark") and semantic negations to assess claim validity. It does not utilize dangerous functions like eval() or exec(), nor does it perform network requests or access sensitive system directories. All file operations are restricted to the workspace directory. The dynamic modification of sys.path is limited to locating internal tooling within the local repository and does not pose a security risk.

evidence-auditor