paper-writing
Pass
Audited by Gen Agent Trust Hub on Nov 1, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCREDENTIALS_UNSAFE
Full Analysis
- [COMMAND_EXECUTION]: The skill dynamically resolves and executes local scripts using bash and python3. Specifically, it executes
extract_paper_style.pyandverify_paper_audits.shfrom paths determined at runtime based on the presence of.aris/tools/or theARIS_REPOenvironment variable. - [INDIRECT_PROMPT_INJECTION]: The pipeline ingests untrusted external data from user-provided reports (
NARRATIVE_REPORT.md) and external style references (URLs, arXiv IDs, or PDFs). This content is interpolated into LLM prompts for planning, writing, and reviewing sections, creating a risk where malicious content in the source data could influence agent behavior. - [EXTERNAL_DOWNLOADS]: The workflow interacts with external sources and APIs, including arXiv for reference lookups and style extraction, web URLs for research, and the Gemini API for figure generation.
- [CREDENTIALS_UNSAFE]: The skill documentation indicates a dependency on the
GEMINI_API_KEYenvironment variable for thegeminiillustration mode. While keys are not hardcoded, the skill requires access to these secrets for full functionality.
Audit Metadata