agentation-self-driving

Pass

Audited by Gen Agent Trust Hub on Mar 6, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill utilizes the agent-browser CLI tool to perform web automation tasks. It constructs and executes shell commands for opening URLs, taking snapshots, and simulating mouse interactions. These commands include placeholders like <url> which could be sensitive to shell injection if the inputs are not properly sanitized by the calling environment.
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8) as it autonomously critiques and interacts with external web page content. A malicious actor could embed instructions within a page's HTML or metadata to influence the agent's behavior.
  • Ingestion points: Untrusted data is ingested from web pages via agent-browser snapshot -i and the results of agent-browser eval calls.
  • Boundary markers: No specific delimiters or "ignore instructions" warnings are utilized when the agent processes the retrieved page content.
  • Capability inventory: The skill has broad capabilities including full browser interaction (clicks, typing, navigation) and JavaScript execution via the agent-browser tool.
  • Sanitization: There is no evidence of sanitization or validation of the page content before it is processed by the agent to determine its next actions.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 6, 2026, 06:38 AM