page-agent
Pass
Audited by Gen Agent Trust Hub on Sep 5, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill fetches the page-agent library and related assets from Alibaba's official GitHub repository and the jsDelivr CDN.
- [COMMAND_EXECUTION]: Provides instructions for standard development operations including cloning the source repository with git and managing dependencies using npm (install, ci, start).
- [DYNAMIC_EXECUTION]: Documentation includes bookmarklet examples that dynamically inject and execute JavaScript code from external sources (CDN or local development server) into the browser context to load the agent panel.
- [INDIRECT_PROMPT_INJECTION]: The skill creates a surface for indirect prompt injection as the agent is designed to read and act upon content within the webpage DOM, which may contain untrusted data.
- Ingestion points: The agent reads the entire webpage DOM as text to interpret the user interface (SKILL.md).
- Boundary markers: The provided instructions do not include specific delimiters or 'ignore' instructions for the agent, relying instead on configuration hooks.
- Capability inventory: The agent possesses the capability to perform DOM operations such as clicking elements and filling form fields based on processed content (SKILL.md).
- Sanitization: The skill explicitly mentions the availability of allowlist and data-masking hooks in the library to mitigate risks associated with processing sensitive or untrusted page content.
Audit Metadata