paper-referee-revise

Pass

Audited by Gen Agent Trust Hub on Jun 22, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses the Bash tool to execute system commands such as pdflatex or xelatex to compile generated LaTeX documents including empirical-revise.tex and response-rX.tex (Phase 3 and Phase 5).
  • [PROMPT_INJECTION]: The skill ingests untrusted text from external referee reports and annotated manuscripts, creating a vulnerability surface for indirect prompt injection. Ingestion points: In Phase 1 and Phase 2, the skill reads files matching patterns like referee*, report*, review*, and annotated* using the Read tool. Boundary markers: Absent; the instructions do not specify the use of delimiters or warnings to the agent to ignore instructions embedded within the report content. Capability inventory: The skill possesses Edit, Write, and Bash capabilities, allowing it to modify files and execute shell commands within the project directory. Sanitization: No validation or sanitization of the extracted referee comments is performed before they are used to generate manuscript revisions. Mitigation: The skill follows a strict iteration loop in Phase 4 that requires explicit user confirmation via the AskUserQuestion tool before any proposed revision is applied to the manuscript, ensuring human oversight of the agent's actions.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 22, 2026, 03:11 AM
Security Audit — agent-trust-hub — paper-referee-revise