The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill documents patterns for processing real-time messages from external PubNub channels, which represents an ingestion surface for untrusted data. 1. Ingestion points: Data enters the agent context via pubnub.addListener and pubnub.fetchMessages (documented in references/dedup-on-merge.md). 2. Boundary markers: The skill instructions recommend using structured JSON envelopes for messages, which defines data boundaries but does not include explicit delimiters for natural language instructions. 3. Capability inventory: Implementation patterns include network communication (pubnub.publish, pubnub.reconnect) and persistent storage writes (IndexedDB, localStorage, SQLite). 4. Sanitization: The patterns emphasize structural validation through schema versioning but do not provide specific sanitization for natural language content within payloads.
[COMMAND_EXECUTION]: The reconnect controller logic in references/backoff-and-jitter.md triggers the pubnub.reconnect() SDK method. This is a standard and expected use of the library to maintain session connectivity.

pubnub-reliability