ml-paper-writing
Warn
Audited by Gen Agent Trust Hub on Feb 17, 2026
Risk Level: MEDIUMEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- EXTERNAL_DOWNLOADS (MEDIUM): The
templates/neurips2025/Makefilecontains anupgradetarget that usescurlto download a ZIP archive frommedia.neurips.ccand subsequently extracts it usingunzip. This represents an unverifiable external dependency that is not on the trusted source list. - COMMAND_EXECUTION (LOW): The Makefile executes several shell commands (
find,xargs,rm,pdflatex,bibtex) to manage build artifacts and compile documents. While these are standard build tools, they represent a capability surface that could be exploited if malicious inputs are provided. - PROMPT_INJECTION (LOW): The skill is vulnerable to indirect prompt injection. LaTeX is a complex system that can execute arbitrary commands (e.g., via
\write18) or include malicious packages. 1. Ingestion points: LaTeX source files (.tex) in the templates directory. 2. Boundary markers: Absent. 3. Capability inventory: Subprocess execution ofpdflatexand file system access via Makefile. 4. Sanitization: Absent.
Audit Metadata