ml-paper-writing
Pass
Audited by Gen Agent Trust Hub on Sep 9, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill fetches metadata and conference style files from well-known academic services.
- Instructions in
SKILL.mduse therequestslibrary to fetch BibTeX metadata fromdoi.org. - The
Makefileintemplates/neurips2025/usescurlto download official style packages frommedia.neurips.cc. - [COMMAND_EXECUTION]: The skill provides templates and logic for compiling LaTeX documents using standard tooling.
- Compilation workflows described in
templates/README.mdandtemplates/neurips2025/Makefileexecutepdflatex,bibtex,latexmk, andpdfcrop. - [INDIRECT_PROMPT_INJECTION]: The skill explores and processes content from research repositories, which represents an attack surface for indirect prompt injection.
- Ingestion points:
SKILL.mdcontains instructions for the agent to search and read contents from a local research repository (e.g.,find,grep,xargs). - Boundary markers: The skill does not implement delimiters or explicit warnings to ignore embedded instructions when reading repository artifacts.
- Capability inventory: The agent has network access (via
requestsand listed dependencies) and file system read access to perform its primary tasks. - Sanitization: No sanitization or filtering is performed on data read from the research repository before it is interpolated into the agent context.
Audit Metadata