site-discovery

Pass

Audited by Gen Agent Trust Hub on Feb 26, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill provides templates for executing shell commands to invoke the MCP client script located at '.claude/skills/mcp-client/scripts/mcp_client.py'. These commands involve passing complex JavaScript code as string arguments to the 'browser_run_code' function, which poses an injection risk if user-controlled data (such as URLs or selectors) is interpolated without robust escaping.\n- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8) because it retrieves and processes content from arbitrary, untrusted websites.\n
  • Ingestion points: Data is ingested from external websites via 'page.goto()' and subsequent extraction logic for links, forms, and accessibility snapshots in 'SKILL.md'.\n
  • Boundary markers: The skill does not implement delimiters or safety instructions to distinguish untrusted web content from agent directives.\n
  • Capability inventory: The skill's execution environment allows for subprocess calls and arbitrary JavaScript execution within a browser session.\n
  • Sanitization: Website data is ingested and processed in its raw form without sanitization or validation.
Audit Metadata
Risk Level
SAFE
Analyzed
Feb 26, 2026, 08:56 PM