customware-support-widget
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEREMOTE_CODE_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The skill instructs the agent to include an external script from
https://app.customware.ai/support-widget/customware-chat.jswithin the application's root component. This script is executed to load the chat widget runtime. As the domain belongs to the skill's author, this is identified as a vendor-provided resource for its intended functionality. - [INDIRECT_PROMPT_INJECTION]: The widget includes a 'page operation' feature that allows it to interact with the host application UI (e.g., clicking buttons, filling forms) in response to user messages. This introduces a capability where external input can influence the host page.
- Ingestion points: The environment receives untrusted input via user messages entered into the widget's chat interface.
- Boundary markers: The widget logic is designed to exclude its own Shadow DOM from operations, and instructions recommend using semantic labels for host elements to ensure intended behavior.
- Capability inventory: The widget can perform actions such as clicking, filling, and submitting elements within the host DOM.
- Sanitization: The instructions explicitly forbid passing sensitive data like API tokens, session cookies, or secrets to the widget, restricting identity metadata to optional name and email fields.
Audit Metadata