What problem does it solve? Application LLM calls often end up with prompts buried in services, reused tracing scenarios, and untracked prompt versions, which corrupts latency, cost, and quality data. This Skill enforces a consistent structure for prompt ownership, versioning, scenario semantics, model policy, and structured output validation. ## Core Features & Use Cases - Prompt Ownership and Versioning: Places reusable generation contracts in packages/prompts with co-located *_PROMPT_VERSION constants formatted as v<major> or v<major>.<minor>. - Scenario and Tracing Discipline: Guides correct use of TRACING_SCENARIOS, schemaName, and entity IDs so each business workflow gets its own tracing cohort. - Model Policy and Structured Output: Resolves models through the owning service's configuration and validates generated JSON against Zod schemas at the service boundary. - Use Case: When adding a new goal-criteria drafting feature, use this Skill to create a versioned prompt chain, register a distinct tracing scenario, configure the service model, and write tests asserting scenario, promptVersion, and schemaName. ## Quick Start Use the llm-generation skill to implement a new structured generation workflow with a versioned prompt, correct tracing scenario, and validation tests.
