The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill documents methods for creating chat bots and slash commands that process untrusted input from messenger users.
Ingestion points: The onMessageAdd and onCommand static methods in rules/bots.md process user-supplied messages and command parameters (COMMAND_PARAMS).
Boundary markers: The provided documentation and code snippets do not include instructions or examples for implementing boundary markers or delimiters to isolate untrusted user data.
Capability inventory: The skill provides full messaging and chat management capabilities, including Bot::addMessage, Chat::sendMessage, and CIMChat::AddMessage (which can bypass user checks if SYSTEM or SKIP_USER_CHECK is enabled).
Sanitization: The example code in rules/bots.md demonstrates direct interpolation of raw user parameters into bot responses ('Report: ' . $fields['COMMAND_PARAMS']) without explicitly showing sanitization or escaping mechanisms.

bitrix24-im