The Agent Skills Directory

[PROMPT_INJECTION]: The skill defines a system for processing untrusted input from external chat platforms (e.g., Slack, Discord, Telegram) through event handlers. This introduces a surface for indirect prompt injection attacks if message content is passed to an AI model or sensitive logic without proper sanitization.
Ingestion points: Handlers such as onNewMention, onDirectMessage, onSubscribedMessage, and onNewMessage in SKILL.md ingest message content from external sources.
Boundary markers: The provided code examples do not include explicit boundary markers or instructions for the agent to ignore embedded commands within user messages.
Capability inventory: The skill utilizes thread.post() and thread.stream() in SKILL.md to respond based on the ingested content.
Sanitization: No input sanitization or validation logic is demonstrated in the implementation examples.
[EXTERNAL_DOWNLOADS]: The skill references a wide range of NPM packages required for the framework and its various adapters. These include the core chat package and specialized adapters like @chat-adapter/slack, @chat-adapter/discord, and @chat-adapter/state-redis. These are standard dependencies for the functionality described.

chat-sdk