nature-reviewer
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill implements a strictly defined workflow for academic manuscript review, emphasizing groundedness, mutual blindness between simulated reviewers, and non-invention of facts.
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input (manuscripts and abstracts), which is a common surface for indirect prompt injection. However, the skill lacks access to sensitive capabilities like network operations or shell execution, and the instructions require the agent to ground all comments in specific evidence from the provided text, which significantly mitigates the risk.
- [COMMAND_EXECUTION]: A Python test script (
tests/test_reviewer_instruction_contracts.py) is included in the package. This script performs static string checks on the skill's instructions and does not execute arbitrary code or interact with external systems.
Audit Metadata