Skip to main content

Interface: JudgeNodeConfig

Defined in: packages/agentos/src/core/llm/providers/judge-config.ts:88

Judge configuration for orchestration human-node delegation — the shared shape for the IR node type, the public builder, and the NodeExecutor runtime (previously three drifting inline copies).

Extends​

  • JudgeLlmConfig

Properties​

confidenceThreshold?​

optional confidenceThreshold: number

Defined in: packages/agentos/src/core/llm/providers/judge-config.ts:92

Confidence threshold (0–1). Below this, fall through to the human interrupt. Defaults to 0.7.

criteria?​

optional criteria: string

Defined in: packages/agentos/src/core/llm/providers/judge-config.ts:90

Custom evaluation criteria/rubric.

effort?​

optional effort: "low" | "medium" | "high" | "xhigh" | "max"

Defined in: packages/agentos/src/core/llm/providers/judge-config.ts:21

Reasoning effort. When omitted, the resolver's default effort is applied ONLY when the resolver's model was also selected — a caller-pinned model gets no injected effort (zero request-shape change for pinned callers).

Inherited from​

JudgeLlmConfig.effort

model?​

optional model: string

Defined in: packages/agentos/src/core/llm/providers/judge-config.ts:13

Judge model id. Default: the central resolver (gpt-5.6).

Inherited from​

JudgeLlmConfig.model

provider?​

optional provider: string

Defined in: packages/agentos/src/core/llm/providers/judge-config.ts:15

Provider id. Pinning a non-openai provider REQUIRES an explicit model.

Inherited from​

JudgeLlmConfig.provider