Class AgentDecisionService

Asks decision questions for an agent run through AIDecisionRunner and the configured decision prompt.

Index

Constructors

Properties

DEFAULT_PROMPT_NAME

Methods

Ask SummarizeAnswers ToDecisionQuestions

Constructors

constructor

new AgentDecisionService(): AgentDecisionService
Returns AgentDecisionService

Properties

`Static` `Readonly`DEFAULT_PROMPT_NAME

DEFAULT_PROMPT_NAME: "Default Decision" = 'Default Decision'

The decision prompt used when none is configured.

Methods

Ask

Ask(args: AgentDecisionAskParams): Promise<AIDecisionRunResult>
Evaluates decision questions against the provided state using the configured decision prompt. Never throws: on failure or missing prompt, returns a failed AIDecisionRunResult.
Parameters
- args: AgentDecisionAskParams
Returns Promise<AIDecisionRunResult>
- Defined in packages/AI/Agents/src/AgentDecisionService.ts:104

`Static`SummarizeAnswers

SummarizeAnswers(
answers: Record<string, DecisionAnswer>,
): Record<string, AgentDecisionAnswerSummary>
Summarises answers for the model: probability, or value and confidence.
Parameters
- answers: Record<string, DecisionAnswer>
Returns Record<string, AgentDecisionAnswerSummary>
- Defined in packages/AI/Agents/src/AgentDecisionService.ts:202

`Static`ToDecisionQuestions

ToDecisionQuestions(questions: Record<string, unknown>): DecisionQuestionMapping
Checks each question's shape and maps the valid ones from the LLM-facing shape (camelCase) onto BaseDecision's (PascalCase). The questions are unchecked model output, so an invalid one is dropped, with its reason, rather than failing the others.
Parameters
- questions: Record<string, unknown>
Returns DecisionQuestionMapping
- Defined in packages/AI/Agents/src/AgentDecisionService.ts:143