paper-writer
Pass
Audited by Gen Agent Trust Hub on Jun 18, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATION
Full Analysis
- [COMMAND_EXECUTION]: The skill provides several local utility scripts for data analysis (
analysis-template.py,table1.py,forest-plot.py) and manuscript compilation (compile-manuscript.sh,word-count.sh). These scripts use standard scientific libraries and operate on local files provided by the user. - [EXTERNAL_DOWNLOADS]: Phase 0 and Phase 1 of the workflow involve using
WebSearchandWebFetchto retrieve journal submission guidelines and literature from established academic databases like PubMed and Google Scholar. This is expected behavior for a research tool. - [DATA_EXFILTRATION]: While the skill manages research directories that may contain sensitive medical data, it includes explicit security instructions in
SKILL.mdanddata-management.mdto de-identify patient data and use.gitignoreto prevent raw data from being committed to version control. No patterns of unauthorized network transmission were found. - [PROMPT_INJECTION]: The skill processes untrusted external data (literature titles and abstracts) which presents an indirect injection surface. However, this is mitigated by the structured 'Literature Matrix' phase and the requirement for human validation of all AI-suggested content.
- [SAFE]: Instructions in
references/hook-compatibility.mdprovide legitimate configuration advice for users whose environments might block the creation of markdown files, ensuring the skill can function as intended without compromising system integrity.
Audit Metadata