qinyan-nature-review
Pass
Audited by Gen Agent Trust Hub on Jul 22, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill is configured to execute a bundled validation script located at
scripts/review_consistency.py. The instructions inSKILL.mddirect the agent to runpython scripts/review_consistency.py <report.md>to verify that generated reports adhere to mandated structural requirements. - [PROMPT_INJECTION]: The skill ingests untrusted manuscript data, representing a surface for indirect prompt injection.
- Ingestion points: The skill's primary input is a user-provided manuscript (referenced in
SKILL.md). - Boundary markers: The skill employs structured report sections (Review setup, Lens A/B/C, Cross-review synthesis) and explicit 'Red Line' prohibitions to constrain agent behavior.
- Capability inventory: The agent possesses capabilities to write files and execute the local bundled Python script.
- Sanitization: No explicit sanitization or filtering of manuscript content is mentioned or implemented in the skill's logic.
Audit Metadata