citation-audit

Pass

Audited by Gen Agent Trust Hub on Nov 12, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface where untrusted data from the manuscript is processed and sent to a high-capability reviewer model.\n
  • Ingestion points: The skill reads *.tex files (Step 1 and Step 2) to extract citation keys and their surrounding sentences for verification.\n
  • Boundary markers: The prompt template provided in Step 3 for the mcp__codex__codex tool does not utilize explicit boundary markers or instructions to ignore potentially malicious content embedded within the extracted LaTeX context.\n
  • Capability inventory: The skill possesses a broad set of capabilities, including Bash(*), Edit, Write, WebSearch, and WebFetch, which could be exploited if the reviewer model's logic is subverted via prompt injection.\n
  • Sanitization: There is no explicit sanitization or escaping layer mentioned for the extracted context strings before they are injected into the reviewer's prompt.\n- [COMMAND_EXECUTION]: The skill employs the Bash(*) tool to perform shell operations, specifically to clean and compile LaTeX projects using latexmk (Step 7). While these commands are standard for document preparation, they represent a significant capability within the shell environment.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 12, 2026, 06:08 AM