The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-supplied data from fields like short_description and description to generate ML-based classifications and recommendations.
Ingestion points: Data enters the agent context via snow_query_table and the predictor.predict(gr) calls which read from the incident table.
Boundary markers: The provided scripts do not include specific delimiters or instructions to ignore embedded prompts within the record fields during ML processing.
Capability inventory: The skill uses snow_execute_script to run server-side JavaScript and snow_query_table to modify record data based on ML outputs.
Sanitization: There is no explicit sanitization or filtering of the input strings before they are passed to the ML prediction engine.
[COMMAND_EXECUTION]: The skill uses snow_execute_script, which allows for the execution of arbitrary server-side JavaScript (ES5) within the ServiceNow environment. While this is a standard administrative tool for the platform, it represents a powerful capability that could be misused if the agent is not properly constrained.

predictive-intelligence