frames
Pass
Audited by Gen Agent Trust Hub on Oct 30, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests and processes untrusted data from external websites during the performance tracing process, creating a potential surface for indirect instructions to influence agent behavior.\n
- Ingestion points: The skill utilizes the
observe.jsutility to extract DOM element descriptors (tag names, class lists, and IDs) and script execution metadata (source URLs, function names, and invokers) from the browser environment.\n - Boundary markers: The skill does not implement explicit delimiters or specific instructions to the agent to disregard potential commands or malicious patterns embedded within the collected page data.\n
- Capability inventory: The agent is tasked with analyzing the gathered performance data to identify bottlenecks and suggest code fixes; the skill also involves browser-level capabilities such as CPU throttling and JavaScript evaluation calls.\n
- Sanitization: There is no evidence of sanitization, filtering, or validation performed on the DOM attributes or script information retrieved from the web page before the data is processed by the agent.
Audit Metadata