The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted external data which serves as a potential ingestion point for indirect prompt injection attacks.
Ingestion points: The skill primarily ingests untrusted data through request.message, request.path, and request.params in various trigger types (Before Publish, After Publish, On Request) across files like enrich-message.js and auth-gateway.js.
Boundary markers: While the skill uses period delimiters for channel patterns, it does not explicitly define LLM boundary markers for embedded data.
Capability inventory: Function scripts possess capabilities for network requests (xhr.fetch), persistent storage writes (kvstore.set), and message publishing (pubnub.publish/fire/signal).
Sanitization: The skill implements validation logic (e.g., checking for required fields) and provides a specific pattern for content moderation using external APIs (Pattern 3 in functions-patterns.md) to sanitize user-generated content.
[COMMAND_EXECUTION]: The bundling and TypeScript reference guide includes shell script snippets for local build tasks, such as bundle size validation and file-based post-processing (fix-export.mjs). These commands are intended for the user's local development environment to support the serverless deployment workflow and are not executed within the agent's runtime environment.

pubnub-functions