skills/zoom/skills/zoom-cobrowse-sdk/Gen Agent Trust Hub

zoom-cobrowse-sdk

Pass

Audited by Gen Agent Trust Hub on Sep 21, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCREDENTIALS_UNSAFEDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill provides instructions to load the Zoom Cobrowse SDK from Zoom's official content delivery network (us01-zcb.zoom.us). This is a standard and expected integration pattern for this service and utilizes a well-known, trusted source.\n- [CREDENTIALS_UNSAFE]: The documentation adheres to security best practices by explicitly instructing developers to generate JWTs in a secure backend environment and strictly warns against exposing the SDK Secret in client-side code.\n- [DATA_EXFILTRATION]: Built-in privacy features are documented, including the piiMask settings and data-cobrowse-mask attributes. These enable developers to sanitize customer browser data and prevent sensitive fields (like SSNs or credit card numbers) from being shared with or viewed by support agents.\n- [INDIRECT_PROMPT_INJECTION]: As a tool designed for real-time collaboration on web pages, the agent's view is derived from the customer's browser content, which represents a potential ingestion surface for untrusted data.\n
  • Ingestion points: The shared browser session content (HTML/DOM) as described in SKILL.md and examples/customer-integration.md.\n
  • Boundary markers: The documentation recommends using the piiMask configuration and CSS selectors to establish data boundaries and prevent unauthorized access to sensitive fields.\n
  • Capability inventory: Agent capabilities include viewing the customer's screen, using annotation tools, and remote scrolling (remoteAssist) as detailed in SKILL.md and examples/remote-assist.md.\n
  • Sanitization: The SDK synchronizes DOM updates and applies masking of sensitive fields as configured by the developer to sanitize the data visible to the agent.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 21, 2026, 10:24 AM