The model proposes. The agent decides.

LLM outputs are trusted by default

Every system that integrates a large language model treats its outputs as candidates for direct use. The model generates text, code, plans, or actions, and the system consumes them — sometimes with filtering, sometimes without. The interface is bidirectional: the model both receives context and produces output that shapes system behavior. This means the model can influence its own future inputs, creating feedback loops that no external filter can fully govern.

Hallucination — the generation of plausible but false content — is not a bug to be fixed. It is a structural property of probabilistic language models. No amount of fine-tuning, RLHF, or prompt engineering eliminates it. The question is not whether the model will hallucinate. The question is what happens when it does.

LLM skill gating provides a unidirectional interface: the model proposes, the agent evaluates, and only structurally validated proposals are accepted. Five structural starvation constraints prevent the model from influencing its own inputs, accumulating unearned authority, or bypassing evaluation. Curriculum-based skill gating governs capability unlocking: the agent progressively gains access to more powerful capabilities only through demonstrated evidence of competent use at each level.

Progressive trust, not blanket authorization

Curriculum-based gating treats capability access as something earned, not granted. A new agent starts with restricted capabilities. As it demonstrates competent use — measured by structural performance metrics, not self-reports — additional capabilities are unlocked. Each gate requires evidence. Each capability has defined prerequisites. Each progression is recorded in the agent's lineage.

This produces agents whose capability scope matches their demonstrated competence. An agent that has not proven it can safely handle simple tasks does not gain access to complex ones. An agent that demonstrates degraded performance has capabilities revoked until performance recovers. The curriculum is the governance — not a separate system applied on top.