academic-paper
Pass
Audited by Gen Agent Trust Hub on Nov 17, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process a wide variety of untrusted third-party data, including pasted reviewer comments, fetched web pages, and uploaded PDFs. This establishes a significant attack surface for indirect prompt injection. However, the skill explicitly mitigates this through a mandatory 'instruction-data-boundary' principle and the use of 'IRON RULE' markers, instructing agents to treat external text exclusively as data to be analyzed rather than commands to be followed.
- [EXTERNAL_DOWNLOADS]: The skill utilizes well-known, trusted external tools and libraries to perform its primary functions. This includes the use of Pandoc for document conversion, Tectonic for LaTeX compilation, and established scientific Python libraries such as Matplotlib and Seaborn for data visualization. These dependencies are from reputable sources and are necessary for the skill's stated academic purpose.
- [COMMAND_EXECUTION]: The orchestration of the pipeline involves executing several local Python scripts (e.g.,
ars_apply_revision_patch.py,check_acronyms.py,verify_submission_package.py) to handle manuscript revisions and quality checks. These scripts provide deterministic management of the document state and ensure adherence to academic standards throughout the multi-phase workflow.
Audit Metadata