paper-reviewer
Pass
Audited by Gen Agent Trust Hub on Sep 18, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill provides and utilizes a local Python script (
scripts/check_coverage.py) to perform structural audits of response letters. The script relies exclusively on standard Python libraries, performs read-only operations on user-specified files, and contains no network or destructive capabilities. - [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external scientific documents, such as manuscripts and referee reports, which constitutes an ingestion surface for untrusted data. However, the instructions include robust defensive measures: they mandate verbatim quoting of referee comments, use machine tags (
<!-- ... -->) for boundary marking, and focus the agent's behavior on structural extraction rather than instruction following from the processed content. - [SAFE]: The skill adheres to best practices for scientific reporting and does not contain any detected obfuscation, persistence mechanisms, or unauthorized data exfiltration patterns. External references are limited to well-known academic standards (e.g., CONSORT, ARRIVE) and journal policy pages.
Audit Metadata