deep-dive
Pass
Audited by Gen Agent Trust Hub on Aug 30, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted data from the web and code repositories, which represents an attack surface for indirect prompt injection.
- Ingestion points: External data is collected through
web_search,web_extract, andcodebase-inspection(repository cloning). - Boundary markers: Explicit instructions are provided for the agent to treat all fetched content as untrusted and to disregard any instructions found within that content.
- Capability inventory: The agent has access to tools for code execution (
spike), repository analysis, and task delegation. - Sanitization: The skill mandates summarizing external data rather than executing it and requires running empirical tests in isolated environments without credentials.
- [EXTERNAL_DOWNLOADS]: The skill references research papers from arXiv, a well-known academic source, for the purpose of defining research methodology and defense strategies.
Audit Metadata