benchmark-fp-fn-audit
Pass
Audited by Gen Agent Trust Hub on Aug 3, 2026
Risk Level: SAFE
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to use
rg --files(ripgrep) to enumerate trial artifacts and discover logs. This is a standard and safe file search operation used for inventorying the local dataset. - [EXTERNAL_DOWNLOADS]: The skill references
https://www.react.doctor/docs/rulesfor documentation. This is a well-known documentation site related to the tool being audited and does not involve executable code downloads. - [DATA_EXFILTRATION]: While the skill reads sensitive-looking paths (e.g.,
/home/aidenybai/Developer/...), all outputs (Markdown, JSONL, TSV) are explicitly restricted to the localaudit/directory within the benchmark path. No instructions involve transmitting this data to external network endpoints. - [SAFE]: The skill follows least-privilege principles by explicitly instructing the agent not to edit React Doctor source code, benchmark source, task tests, or verifier data, ensuring the audit process remains non-destructive.
Audit Metadata