read-arxiv-paper

Pass

Audited by Gen Agent Trust Hub on Feb 17, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS] (LOW): The skill downloads compressed .tar.gz archives from arxiv.org. Although the source is a standard academic site, the content is unverified and provided by third parties.
  • [PROMPT_INJECTION] (LOW): The skill is susceptible to Indirect Prompt Injection (Category 8).
  • Ingestion points: LaTeX source files downloaded and unpacked into ~/.cache/nanochat/knowledge/.
  • Boundary markers: Absent. There are no instructions or delimiters provided to the agent to treat the LaTeX content as data rather than instructions.
  • Capability inventory: The agent has the ability to read the entire local nanochat repository and write markdown files to the local ./knowledge/ directory.
  • Sanitization: Absent. There is no validation or filtering of the LaTeX content before the agent reads and processes it.
Audit Metadata
Risk Level
SAFE
Analyzed
Feb 17, 2026, 04:53 PM