alphaxiv

Pass

Audited by Gen Agent Trust Hub on Nov 10, 2026

Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes a bash script in Step 6 that interpolates values derived from external content directly into a shell command. While intended for integration with the 'research-wiki' skill, this pattern creates a risk of command injection if the input strings are maliciously crafted. Evidence: python3 "$WIKI_SCRIPT" ingest_paper ... --arxiv-id "<paper_arxiv_id>" [--thesis ""] in SKILL.md.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external content from AlphaXiv and arXiv without explicit boundary markers or sanitization, potentially allowing embedded instructions in academic papers to influence agent behavior.
  • Ingestion points: Markdown and LaTeX content fetched via curl from alphaxiv.org and arxiv.org (Steps 2, 3, and 4).
  • Boundary markers: None observed; the skill instructions do not specify delimiters for the fetched content.
  • Capability inventory: Access to Bash, Read, Write, and Glob tools.
  • Sanitization: No validation or filtering of the fetched text is performed before it is summarized or passed to other tools.
  • [EXTERNAL_DOWNLOADS]: Fetches paper summaries from AlphaXiv and source archives from arXiv. Evidence: Performs curl requests to alphaxiv.org and arxiv.org in Steps 2, 3, and 4. These are established academic platforms.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 10, 2026, 09:36 AM