The Agent Skills Directory

Indirect Prompt Injection Surface: The skill processes user-provided business definitions and outcome descriptions via the AskUserQuestion tool. This ingestion of untrusted data represents a potential surface for indirect instructions. The skill addresses this by mandating confirmation of entity definitions and provenance, and by requiring validation through the validate_custom_metric_formula tool. Evidence: (1) Ingestion points: AskUserQuestion and business descriptions. (2) Boundary markers: Instructions to confirm case/event entity, outcome definition, and provenance. (3) Capability inventory: Read, AskUserQuestion, validate_custom_metric_formula, and get_cases_with_metrics_v2. (4) Sanitization: Mandatory use of the validate_custom_metric_formula tool and a limit of three validation attempts.
Dynamic Formula Execution: The skill enables the evaluation of ephemeral formulas at runtime. This dynamic execution is managed by the validate_custom_metric_formula tool, which checks formulas against specific process metadata (processVersionId, processExtendedMetadataVersionId). The skill also limits validation attempts to three to prevent resource exhaustion or brute-forcing during logic repair.

derive-metric