light-paper-drafting
Pass
Audited by Gen Agent Trust Hub on Jun 23, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill utilizes command-line tools as part of its documented workflow, specifically the 'markitdown' utility for document conversion and the 'scripts/draft_lint.py' Python script for validating draft integrity, checking for AI failure modes, and identifying missing sections.
- [EXTERNAL_DOWNLOADS]: The documentation references the installation of 'markitdown' (a well-known service from Microsoft) and provides 'curl' commands for querying academic metadata services such as DOI.org, Crossref, arXiv, and OpenAlex. These network operations target well-known repositories for the legitimate purpose of citation verification and are documented neutrally.
- [PROMPT_INJECTION]: The skill contains an attack surface for indirect prompt injection because it is designed to ingest and process external research papers (PDF, DOCX, PPTX) converted to text via 'markitdown'. This could allow malicious instructions embedded in external papers to influence the agent.
- Ingestion points: External document files processed through the 'markitdown' converter and fed into the agent context.
- Boundary markers: The skill employs instructional markers like '[MATERIAL GAP]' for content tracking but does not use formal security delimiters to separate external data from system instructions.
- Capability inventory: The skill has access to shell execution for running conversion and linting tools, as well as file system read/write capabilities for draft management.
- Sanitization: No explicit sanitization or filtering of the text extracted from external research documents is performed prior to processing.
Audit Metadata