Assurance source reference
Generative AI Assurance
Activate cross-cutting controls for models, prompts, RAG, tools, memory, agents, evals, and AI lifecycle risk.
Detect
AI assurance applies when a model, provider, prompt, embedding, retrieval corpus, AI tool, AI memory, or agent can affect workload behavior.
Route
Use valdris-genai-assurance as primary for AI-centered work or supporting for an AI-bearing feature, security audit, architecture change, or release.
Prove
Write ai/assurance.json, set rag, tools, and memory feature flags explicitly, and prove all active catalog controls. Also write and validate evals/results.json and trajectory/trajectory.json.
Red Zone
Human approval remains required for consequential tools, sensitive AI data, provider/model production promotion, weakened safety policy, destructive corpus/memory changes, and production rollout or rollback.
Finish
Run node scripts/enterprise-ai-gate-all.mjs --repo .. Runtime-native completion is advisory until Valdris gates and approvals pass.