browser-auth-flow

Pass

Audited by Gen Agent Trust Hub on Jun 13, 2026

Risk Level: SAFE
Full Analysis
  • [COMMAND_EXECUTION]: The skill instructs the agent to use browser automation tools and Bash to perform security audits. This usage is consistent with its stated purpose of probing authentication flows for vulnerabilities.
  • [DATA_EXFILTRATION]: The skill captures sensitive session information such as cookies and OAuth tokens. However, it explicitly includes a quarantine process where data is stored within a local container and must pass through PII scanning tools (mcp__claude-flow__aidefence_scan) and safety gates before being processed by the model, preventing unauthorized exfiltration.
  • [PROMPT_INJECTION]: As a tool that processes external website content, the skill is naturally exposed to an indirect prompt injection surface.
  • Ingestion points: Data enters the agent context through browser snapshots, URL monitoring, and evaluation of DOM properties (document.cookie) from untrusted external login URLs.
  • Boundary markers: The instructions do not define specific text delimiters for untrusted data, relying instead on the quarantine mechanism.
  • Capability inventory: The skill has access to shell commands via Bash and file writing capabilities to record findings.
  • Sanitization: The skill mitigates risks by requiring PII scanning (aidefence_scan) and redaction gates before returning any captured data to the model context.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 13, 2026, 01:22 PM
Security Audit — agent-trust-hub — browser-auth-flow