agent-browser

Pass

Audited by Gen Agent Trust Hub on Feb 21, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION] (LOW): Indirect Prompt Injection Surface (Category 8).
  • Ingestion points: Untrusted data enters the agent context through agent-browser snapshot, agent-browser get text, agent-browser get title, and agent-browser get url in templates/capture-workflow.sh, templates/form-automation.sh, templates/authenticated-session.sh, and references/authentication.md.
  • Boundary markers: Absent across all provided templates and reference files. There are no delimiters or instructions provided to the agent to treat scraped web content as untrusted.
  • Capability inventory: Significant capabilities are present across the skill scripts, including network navigation (agent-browser open), form interaction (agent-browser fill, agent-browser click), file-system operations (agent-browser state save, agent-browser screenshot, agent-browser pdf in templates/capture-workflow.sh), and file interaction (agent-browser upload in templates/form-automation.sh).
  • Sanitization: Absent. No evidence of escaping, validation, or filtering of content retrieved from the browser before it is processed by the agent.
  • [CREDENTIALS_UNSAFE] (SAFE): No hardcoded credentials were detected. The skill properly advises using environment variables for sensitive data and recommends excluding session state files from version control using .gitignore in references/authentication.md and references/session-management.md.
Audit Metadata
Risk Level
SAFE
Analyzed
Feb 21, 2026, 12:18 AM