Skip to main content

Class: EmergentAgentJudge

Defined in: packages/agentos/src/cognition/emergent/EmergentAgentJudge.ts:90

Strict LLM-as-judge gate for synthesized agent specs.

Constructed once per agency strategy execution; reused across multiple spawn_specialist calls. Stateless aside from the immutable config.

Example​

const judge = new EmergentAgentJudge({
  judgeModel: 'gpt-4o-mini',
  generateText: async (model, prompt) => callLlm(model, prompt),
});

const verdict = await judge.reviewAgent(spec);
if (!verdict.approved) {
  return { success: false, data: `Judge rejected: ${verdict.reason}` };
}

Constructors​

Constructor​

new EmergentAgentJudge(config): EmergentAgentJudge

Defined in: packages/agentos/src/cognition/emergent/EmergentAgentJudge.ts:93

Parameters​

config​

EmergentAgentJudgeConfig

Returns​

EmergentAgentJudge

Methods​

reviewAgent()​

reviewAgent(spec): Promise<AgentVerdict>

Defined in: packages/agentos/src/cognition/emergent/EmergentAgentJudge.ts:102

Evaluate an agent spec. Never throws — all failure paths return a structured rejection so the caller can surface a clean error to the manager LLM.

Parameters​

spec​

AgentSpec

Returns​

Promise<AgentVerdict>