inline-embeds

Pass

Audited by Gen Agent Trust Hub on Jul 13, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [SAFE]: The skill documents the platform's 'embed' code fence, which renders interactive content within sandboxed iframes. These iframes use 'allow-scripts', 'allow-same-origin', 'allow-forms', and 'allow-popups' to isolate the embedded application from the agent's parent context.
  • [EXTERNAL_DOWNLOADS]: The skill references the '@agent-native/core/client' Node.js package, which provides utility functions like 'postNavigate' and 'isInAgentEmbed' for managing the interactive state between the agent and the application.
  • [SAFE]: The 'postNavigate' function implements security checks by validating that 'postMessage' payloads originate from the same origin and the same iframe, restricting navigation to authorized relative paths.
  • [PROMPT_INJECTION]: The skill defines a surface for indirect prompt injection where an agent might be directed to render a specific route with interpolated data. * Ingestion points: Data extracted from the application context, such as message or ticket identifiers, used to construct the 'src' path. * Boundary markers: The 'embed' fenced code block serves as a structural delimiter for the interactive output. * Capability inventory: The agent has the ability to render internal application routes and trigger client-side navigation via platform-provided helpers. * Sanitization: The framework-level renderer automatically enforces same-origin policies, blocks cross-origin URLs, and applies restrictive sandbox attributes to prevent unauthorized access to the parent window context.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 13, 2026, 04:51 AM
Security Audit — agent-trust-hub — inline-embeds