cli-anything-safari

Pass

Audited by Gen Agent Trust Hub on Sep 15, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill triggers the download of the safari-mcp package from the NPM registry during its first tool invocation via npx. NPM is recognized as a well-known package registry.\n- [DYNAMIC_EXECUTION]: The evaluate and run-script tools allow the agent to execute arbitrary JavaScript code within the active browser tab. This is a core feature for web automation but constitutes a dynamic code execution path.\n- [INDIRECT_PROMPT_INJECTION]: As the skill retrieves and processes content from the public internet, it creates an attack surface where maliciously crafted web content could attempt to influence the agent.\n
  • Ingestion points: tool snapshot, tool read-page, and tool navigate-and-read which fetch remote content.\n
  • Boundary markers: The CLI provides structured JSON output for tool responses when requested via the --json flag.\n
  • Capability inventory: Includes JavaScript execution, local file writes (PDF generation), and system clipboard access.\n
  • Sanitization: Employs a security utility to validate URLs and block dangerous protocols like file:// or javascript:.\n- [COMMAND_EXECUTION]: The tool operates by spawning subprocesses to run the Node.js MCP server (npx safari-mcp) for each requested action.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 15, 2026, 05:21 PM
Security Audit — agent-trust-hub — cli-anything-safari