manuscript-review-council

Pass

Audited by Gen Agent Trust Hub on Apr 12, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data (manuscripts in PDF, DOCX, or text format), which creates a surface for indirect prompt injection attacks where malicious instructions hidden in a manuscript could influence the reviewer agents or the final decision.
  • Ingestion points: Untrusted manuscript content is gathered in SKILL.md (Instruction 2) and normalized into a shared packet for reviewers as defined in references/council-workflow.md.
  • Boundary markers: The skill lacks explicit instructions or technical delimiters to ignore or escape instructions that might be embedded within the manuscript text, although it does provide a behavioral guardrail by instructing reviewers to ground criticisms only in the text or explicit missing information (Instruction 9).
  • Capability inventory: The skill utilizes powerful capabilities including agent delegation (spawning sub-agents or Task agents) and specialized analytical tools (/polars-dovmed, /bio-logic, /scientific-writing) to process and act upon the provided content.
  • Sanitization: There is no evidence of sanitization, filtering, or validation of the manuscript content before it is processed by the agent council.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 12, 2026, 02:58 AM