vp-agent-browser-session

Pass

Audited by Gen Agent Trust Hub on Aug 11, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill utilizes a helper script, agent-browser-sessionctl, which spawns the agent-browser CLI tool for session management and browser control.
  • [PROMPT_INJECTION]: The skill incorporates a surface for indirect prompt injection by instructing the agent to follow guidance dynamically retrieved from the local agent-browser tool. Ingestion points: agent-browser CLI output and web content. Boundary markers: Explicit rules in SKILL.md regarding profile isolation and privacy. Capability inventory: File management and CLI execution. Sanitization: Profile name regex validation and argument blocklisting.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 11, 2026, 02:41 AM
Security Audit — agent-trust-hub — vp-agent-browser-session