The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill documentation includes instructions to install the @friqdev/jev-kit and ai packages via package managers. The @friqdev/jev-kit package is associated with the skill author, and ai is a well-known library from Vercel.
[INDIRECT_PROMPT_INJECTION]: The skill describes a workflow that ingests untrusted text data for classification and decision-making purposes.
Ingestion points: Untrusted data enters the agent context through the state object passed to the jev.evaluate function (e.g., subject and body fields in the provided example).
Boundary markers: The skill emphasizes structured output via a QuestionMap but does not define explicit delimiters for the input text, relying on the model's internal classification capabilities.
Capability inventory: Decision results from the model (e.g., urgency, category, severity) are used to determine automated actions, such as routing tickets or triggering alerts.
Sanitization: The skill includes a clear safety warning stating that the model is not a security boundary and will process adversarial or injected content literally, advising users not to use it for intent-gating on untrusted input without additional controls.

jev