What problem does it solve?
Manually creating ground-truth test suites for OCS chatbot deep gate evaluation is time-consuming and often misses critical coverage, especially for adversarial edge cases and archetype-specific question categories that trigger false eval failures.
Core Features & Use Cases
- Archetype-aligned prompt generation: Automatically branches on the PDD's declared archetype (atomic-visit, focus-group, multi-stage) to generate category-specific Q&A pairs that match the opportunity's workflow, eliminating mismatched category sets that cause false-positive eval failures.
- Mandatory adversarial coverage: Enforces inclusion of 7 required adversarial prompt categories (including safety-critical, hallucination-probe, and ambiguous-intent) with a minimum 20% share of the total suite to catch RAG hallucinations, safety gaps, and out-of-scope responses.
- Validated ground-truth output: Produces a standardized markdown test suite with expected answers, tags, and escalation rules for Phase 5 OCS chatbot deep evaluation, with built-in self-coverage checks to ensure the suite meets eval requirements before submission.
- Real-world use case: For a focus-group archetype PDD for a malaria ITN distribution opportunity, this skill generates the full test prompt file covering session flow, consent, facilitation technique, and all required adversarial categories, ready for immediate use in the Phase 5 deep gate.
Quick Start
Use the pdd-to-test-prompts skill to generate the complete ground-truth test suite for your approved PDD for the [opp-name] opportunity, ready for Phase 5 OCS chatbot deep evaluation.
