page-agent
Pass
Audited by Gen Agent Trust Hub on Sep 19, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill demonstrates integration patterns where untrusted data or user-supplied input is directly interpolated into natural language prompts used to control the agent. A malicious input could potentially hijack the agent's logic to perform unintended actions on the webpage.\n
- Ingestion points: User input from input fields (SKILL.md), CRM data records (references/use-cases.md), and speech recognition transcripts (references/use-cases.md) are passed into the agent's execution engine.\n
- Boundary markers: The prompt templates do not use specific delimiters or instructions to isolate variable data from the agent's primary instructions.\n
- Capability inventory: The Page Agent possesses the capability to interact with the Document Object Model (DOM), including clicking buttons, filling forms, and navigating complex UI structures (SKILL.md, references/use-cases.md).\n
- Sanitization: The provided code examples use direct string interpolation (template literals) without visible sanitization, although the documentation explicitly recommends the use of the
onBeforeActionconfirmation hook to mitigate risks during sensitive operations.\n- [EXTERNAL_DOWNLOADS]: The skill documentation facilitates the download and execution of thepage-agentlibrary via NPM and a well-known public CDN. These resources originate from a well-known technology organization's official repository and standard package registries.
Audit Metadata