ai-product-extension

What problem does it solve?

Prevents unsafe, leaking, or low-quality AI behavior by enforcing product-grade engineering rules for LLMs, RAG, agent tool use, evaluation, and production governance.

Core Features & Use Cases

• LLM safety and trust-boundary controls: prompts are injection-aware, outputs are treated as probabilistic, and downstream use validates untrusted model output.
• Permission-aware RAG requirements: retrieval must apply the same ACL/RBAC filters as the source system to avoid cross-tenant leakage.
• Evaluation-driven release gates: requires ground-truth datasets and defines quality thresholds for hallucination, refusal correctness, and adversarial cases.
• MLOps governance for AI lifecycle: enforces model registry versioning, training-serving skew checks, drift monitoring, and rollback readiness.

Quick Start

Apply the ai-product-extension skill to your proposed LLM/RAG/agent change and ask the agent to produce an AI risk assessment plus a release checklist that includes evaluation coverage, permission-filtering requirements, and required safety controls.