Laporan audit skill

stanford-ai-agent-skill Laporan audit.

Use when designing, debugging, training, researching, or evaluating an AI agent and the decision concerns one of seven areas: minimum agent loop, inference-time compute allocation, verifier or LLM-judge trust, long-horizon dependency and rollback control, verified-trajectory post-training, evidence-gap research, or production-readiness evaluation. Trigger on phrases such as "agent architecture", "test-time scaling", "verifier", "judge audit", "rollback", "SFT or RL", "deep research", and "production ready". Do not use for generic AI explanations, model-news lookup, or ordinary software tasks with no agent-control decision.