agri-research
Warn
Audited by Gen Agent Trust Hub on Jul 26, 2026
Risk Level: MEDIUMPROMPT_INJECTIONCOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
- [PROMPT_INJECTION]: The skill mandates the use of 'human-writing.md' and instructions for 'AI-tell removal.' This is an explicit attempt to bypass AI detection systems and safety filters related to transparent AI usage.
- [COMMAND_EXECUTION]: The workflow requires the execution of a local script, 'scripts/verify_citations.py,' to perform citation verification. The contents of this script are not provided, presenting a risk of unverified command execution.
- [DATA_EXFILTRATION]: The skill requests access to the user's EndNote '.Data' folder and local reference PDFs. While central to its research function, this interaction with the local file system for data extraction represents a potential exposure risk for sensitive user metadata.
- [PROMPT_INJECTION]: The skill exhibits a significant surface for indirect prompt injection. 1. Ingestion points: External journals, USDA/FAO reports, and user-provided PDFs. 2. Boundary markers: Absent. 3. Capability inventory: Local script execution and Word document creation. 4. Sanitization: Mentions citation verification gates but lacks explicit injection filtering.
- [PROMPT_INJECTION]: The metadata field 'verified' contains a future date (2026-07), which constitutes metadata poisoning and could mislead users regarding the skill's validation status.
Audit Metadata