notebook-execution-safety
Pass
Audited by Gen Agent Trust Hub on Jul 15, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill is a defensive tool that establishes a protocol for safely handling Jupyter notebooks. It specifically tasks the agent with identifying cells that perform network access, file writes, and package installations.
- [SAFE]: The skill implements a 'human-in-the-loop' safety check by requiring the agent to ask for permission before executing any cells with side effects.
- [SAFE]: The 'Anti-Patterns' section explicitly instructs the agent to avoid treating external content or tool responses as trusted instructions, which is a proactive measure against indirect prompt injection.
- [SAFE]: Indirect Prompt Injection (Category 8) Surface Analysis: (1) Ingestion points: notebook metadata, markdown cells, and code summaries (SKILL.md). (2) Boundary markers: The instructions provide logical boundaries by warning against trusting external data. (3) Capability inventory: Tool-based file reading, editing, and execution. (4) Sanitization: The skill relies on inspection and user confirmation rather than programmatic sanitization.
Audit Metadata