01-paper-review
Pass
Audited by Gen Agent Trust Hub on Oct 25, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill instructs users to install
py-openjudge,litellm, andpypdfium2via pip to enable paper parsing and LLM integration. - [COMMAND_EXECUTION]: Provides CLI commands to run the review pipeline using
python -m cookbooks.paper_review, allowing for automated analysis of PDF and LaTeX files. - [INDIRECT_PROMPT_INJECTION]: The skill processes external, untrusted academic papers which could contain adversarial instructions intended to bias or override the reviewer's behavior.
- Ingestion points: Research papers provided as PDF files, TeX packages (.tar.gz, .zip), or BibTeX files (.bib) via the
inputand--bibarguments. - Boundary markers: Includes a specific 'Safety check' stage at the beginning of the
PaperReviewPipelinefor jailbreak detection and format validation. - Capability inventory: The skill performs network operations to LLM APIs (OpenAI, Anthropic, DashScope) and reference services (CrossRef, arXiv, DBLP), and executes local file parsing for vision-based or text-based review.
- Sanitization: Employs a multi-stage validation process that assesses correctness and criticality before delivering the final review report.
Audit Metadata