The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides instructions for developing applications that ingest data from a host portal and external services.
Ingestion points: Micro-apps receive context and configuration data via an InitPayload (containing user roles, permissions, and routing info) during the SDK handshake and subscribe to real-time updates through event listeners such as sdk.onRoute and sdk.onTheme (documented in SKILL.md and references/sdk-reference.md).
Boundary markers: Applications are isolated within an iframe utilizing specific sandbox attributes (allow-scripts, allow-same-origin) and referrerPolicy. The SDK validates allowedOrigins for all cross-frame communication.
Capability inventory: Applications can trigger platform-level notifications (sdk.notify), update dashboard widgets, and manage scheduled tasks via the provided SDK methods.
Sanitization: The documentation directs developers to use the @xgent/file-preview package for markdown rendering and mandates the use of sdk.callService to proxy API requests securely, reducing direct exposure to external data.
[DATA_EXFILTRATION]: The skill describes network interaction capabilities, including a secure service proxy (sdk.callService) and a specialized download handler (openDownload). These patterns are standard for the micro-app architecture and are documented with security constraints to prevent common browser-based vulnerabilities like CSP violations.
[CREDENTIALS_UNSAFE]: The documentation explicitly promotes secure secret management by instructing developers not to persist authentication tokens (TDT) in local storage or environment variables, instead relying on the SDK's internal caching and automatic renewal mechanisms.

portal-micro-app