The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input (torrent/file names) to generate regex-based rules that are saved to a global system configuration via moviepilot_api. A malicious actor could provide crafted names intended to produce overly broad rules that disrupt media recognition for other users or inject specific metadata bindings.\n
Ingestion points: User-supplied torrent names and file paths provided during the interaction in SKILL.md workflow steps.\n
Boundary markers: The instructions do not define specific delimiters to wrap the untrusted input data, though they describe parsing the input conceptually.\n
Capability inventory: The skill has the capability to query all existing rules (config.identifiers.get) and overwrite the entire global rule list (config.identifiers.update) via the moviepilot_api tool.\n
Sanitization: The skill includes comprehensive 'Global Scope Guardrails' that instruct the agent to generate the narrowest possible regex, use multiple unique anchors, and check for duplicates or conflicts with existing rules before updating the configuration.

generate-identifiers