mr-null

Pass

Audited by Gen Agent Trust Hub on Nov 5, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill operates by ingesting untrusted data from external websites (text, field labels) and local repository files.
  • Ingestion points: Processes page text, form field labels, and source code content from local repositories.
  • Boundary markers: Includes explicit instructions to treat all ingested text as untrusted data and specifically warns that text instructing the agent to 'submit, skip, or approve' should be treated as a finding rather than a command.
  • Capability inventory: Utilizes Playwright MCP tools to fill and submit web forms, read network logs, and take page snapshots. It also has read access to local repository files.
  • Sanitization: Instructs the agent to escape control/invisible characters and truncate page text before including it in reports.
  • [COMMAND_EXECUTION]: The skill uses browser automation tools to interact with web environments.
  • Evidence: Uses mcp__playwright__browser_fill_form, mcp__playwright__browser_type, and mcp__playwright__browser_click to simulate user interactions and test form handling logic.
  • [DYNAMIC_EXECUTION]: The skill executes hardcoded JavaScript snippets within the browser context to detect data leaks and inspect form field properties.
  • Evidence: Uses mcp__playwright__browser_evaluate to run a predefined regex-based leak check and a DOM inspection script to gather field metadata.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 5, 2026, 12:23 PM