The Agent Skills Directory

[SAFE]: The skill consists of documentation and Python code snippets for architectural reference. It does not include any executable scripts, tools, or commands that could compromise the system.
[INDIRECT_PROMPT_INJECTION]: The skill defines a system architecture that ingests untrusted external data in the form of prospect messages (actual_response). This represents a potential attack surface for indirect prompt injection if the final implementation does not sanitize inputs.
Ingestion points: Prospect messages are ingested via the actual_response variable in the select_preloaded_reply and get_reply_for_response functions within SKILL.md.
Boundary markers: No explicit boundary markers or delimiters for user input are present in the illustrative code snippets.
Capability inventory: The provided code contains logic for classification and caching but does not include any capabilities for file system access, network operations, or subprocess execution.
Sanitization: The snippets do not demonstrate input sanitization or validation techniques for the processed messages.

reply-prediction