academic-paper-reviewer
Pass
Audited by Gen Agent Trust Hub on Oct 15, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill demonstrates high security awareness by defining a 'canonical:instruction-data-boundary'. It explicitly instructs the AI agents that external content (pasted text, PDFs, manuscripts) is data and not instructions, even if it contains imperative text.
- [COMMAND_EXECUTION]: The skill uses several internal Python scripts for synthesis, validation, and provenance tracking (e.g.,
check_panel_synthesis.py,review_panel_provenance.py). These are standard operational tools for the skill's workflow and execute in a controlled environment. - [INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted author-supplied manuscripts. It mitigates the risk of indirect prompt injection through the use of specific data delimiters (e.g.,
<paper_content>) and clear meta-instructions telling agents to ignore any 'ignore previous instructions' type triggers found within the data. - [REMOTE_CODE_EXECUTION]: There is no evidence of downloading or executing untrusted remote code. References to external providers (cross-model verification) are gated by explicit user consent requirements.
- [DATA_EXFILTRATION]: No malicious network operations or exfiltration patterns were detected. The skill uses a structured approach to handle external API calls for cross-model verification only after user approval.
Audit Metadata