academic-paper-reviewer
Pass
Audited by Gen Agent Trust Hub on May 21, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill implements a robust multi-agent architecture with clearly defined roles (EIC, methodology, domain, etc.) and strict phase boundaries to maintain integrity during the review process.
- [PROMPT_INJECTION]: Deterministic detections for 'ignore instructions' in agent files (e.g., eic_agent.md, devils_advocate_reviewer_agent.md) are identified as false positives. These are explicitly defensive instructions designed to prevent the LLM from obeying any malicious prompts embedded in the academic papers it is tasked with reviewing (Indirect Prompt Injection mitigation).
- [EXTERNAL_DOWNLOADS]: No remote dependencies or external script downloads are utilized. Referenced Python scripts (e.g., check_pipeline_integrity.py) are internal validation tools described as part of the orchestration flow.
- [DATA_EXFILTRATION]: No suspicious network operations or unauthorized file access patterns were found. The skill documentation explicitly states that calibration data is session-scoped and not persisted in permanent storage.
- [COMMAND_EXECUTION]: The skill uses localized script calls for internal verification of pipeline integrity and contract adherence, but does not provide vectors for arbitrary command execution from user input.
Audit Metadata