Agent
A specialized cognitive worker that produces a proposal, evaluation, prediction, or retrieval result under a shared runtime contract. The cognitive loop coordinates a planner agent, executor agent, critic agent, memory agent, and meta-cognition agent as standard roles.
An agent in the Cognitive Substrate is not a general-purpose AI system - it is a bounded worker with a single clearly defined role in the cognitive loop. The shared runtime contract means every agent receives the same AgentContext (triggering event, retrieved memories, active goals, policy vector), emits its output in the same structured format, and contributes an activity trace to the same index. This uniformity is what allows agents to run concurrently, be scored by arbitration on comparable dimensions, and be replaced or tuned independently without disrupting the rest of the pipeline. The five standard roles - planner, executor, critic, memory, and meta-cognition - cover the full deliberation cycle, but the architecture is extensible: open-ended evolution can propose new agent roles when the existing set proves insufficient for a class of problem. Each role has a distinct retrieval strategy: the planner pulls episodic memories, the executor pulls procedural ones, the critic pulls contradiction records.