paper-self-revise

Pass

Audited by Gen Agent Trust Hub on Jun 22, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill utilizes the Bash tool to perform utility tasks such as word counting (wc -w) and document conversion using pandoc. While these are legitimate for the workflow, they involve executing shell commands on the host environment.
  • [PROMPT_INJECTION]: The skill is susceptible to Indirect Prompt Injection because it reads and processes external, potentially untrusted content from manuscripts (main.tex) and review reports (PDF, DOCX, TXT, MD, TEX).
  • Ingestion points: Files in the user-provided project folder accessed via Read, Glob, and Grep tools.
  • Boundary markers: The instructions lack specific delimiters or guardrails to prevent the agent from obeying instructions hidden within the academic text or reviewer comments.
  • Capability inventory: The combination of Bash, Write, and Edit tools creates a capability surface that could be exploited if the agent is manipulated by malicious data in the input files.
  • Sanitization: There is no mention of sanitizing or escaping the content read from the manuscript or review materials before it is processed or presented.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 22, 2026, 03:11 AM
Security Audit — agent-trust-hub — paper-self-revise