The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input in the form of torrent titles and file names provided by users to generate identification rules. This represents a potential surface for indirect prompt injection where a malicious file name could attempt to influence the agent's rule generation logic. However, the skill provides specific workflows requiring the agent to query existing rules and verify duplicates, which serves as a procedural mitigation.
Ingestion points: User-provided torrent names or file paths (processed during the 'Analyze the Problem' step).
Boundary markers: The skill does not define explicit delimiters for the input data in its instructions to the agent.
Capability inventory: Includes the ability to read existing rules (query_custom_identifiers) and overwrite the entire rule list (update_custom_identifiers), which is a powerful capability.
Sanitization: The instructions remind the agent to escape regex special characters, reducing the risk of accidental rule breakage.
[COMMAND_EXECUTION]: While the skill generates regex rules and identifier patterns, it does not execute arbitrary shell commands. It uses predefined MCP tools to interact with the media recognition system in a structured manner.

generate-identifiers