What problem does it solve? AI model providers ship new versions constantly, causing silent output drift that breaks brand consistency, year-over-year campaign comparability, and legal or regulatory reproducibility requirements. This Skill establishes a versioned registry manifest that pins exact model versions, prompt fingerprints, brand tokens, and seeds so any asset can be regenerated identically months or years later. ## Core Features & Use Cases - Model alias pinning: Map friendly aliases like flux or kling to exact model@version IDs so teams never reference raw versions directly. - Prompt fingerprinting: Store SHA256 hashes of prompt templates so any string change triggers a versioned bump instead of silent drift. - Regression-tested bumps: Run a 10-20 prompt corpus against candidate versions, diff outputs, and get approval before tagging a new registry release. - Use Case: A 20-person design team at a regulated company commits registry.json to a governance repo, configures the CLI with gen-ai config set registry.path, and regenerates last year's annual report graphics identically for a legal hold. ## Quick Start Ask the AI to author a registry.json pinning your model aliases, prompt templates, brand tokens, and default seed, then configure the gen-ai CLI to resolve all generation calls against it.
