manuscript-writing-review
Pass
Audited by Gen Agent Trust Hub on Aug 4, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill consists entirely of natural language instructions and Markdown formatting designed to guide an AI assistant in performing editorial reviews of scientific text. It does not include executable scripts, binary files, or network-enabled tools.
- [DATA_EXFILTRATION]: No evidence of data exfiltration was found. The skill operates on text provided by the user within the agent's context and does not attempt to access sensitive local files or send data to external servers.
- [PROMPT_INJECTION]: The instructions are well-structured and do not contain patterns intended to bypass AI safety guardrails or override system prompts. The 'Severity Levels' mentioned in the skill are internal labels for the writing report it generates for the user, not instructions to the AI to prioritize malicious actions.
- [INDIRECT_PROMPT_INJECTION]: While the skill is designed to process untrusted user manuscripts (Category 8 surface), it lacks the capabilities (such as network access or shell execution) required to exploit such an injection. It is evaluated as SAFE due to the lack of an exploitable capability chain.
Audit Metadata