kill-argument
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from external LaTeX, BibTeX, and PDF files which could contain malicious instructions designed to subvert the agent's behavior.
- Ingestion points: The skill inventories and reads all
.texand.bibfiles, as well as PDFs, within a user-specified directory defined by$ARGUMENTS(Workflow Steps 1 and 2). - Boundary markers: The prompt templates provided to the external
mcp__codex__codextool in Steps 2 and 3 lack explicit delimiters or instructions to ignore embedded prompts within the ingested research files. - Capability inventory: The skill has access to
Bash(*),Write, andEdit, enabling the agent to execute shell commands and modify local files based on the review outcome. - Sanitization: No sanitization, validation, or filtering of the file contents is performed before the data is passed to the reviewer models or processed by shell tools.
Audit Metadata