Member Junction
    Preparing search index...

    Structured input to the judge LLM. Used by LLMJudge and MJLLMJudge to build judge prompts.

    Index

    Constructors

    Properties

    ApplicationContext?: string

    The same free-form application context the controller receives. The judge needs it for the same reason: without it, app-specific facts — how long a boot screen legitimately persists, what a loading state looks like — have to be hardcoded into the shared prompt, where they ship to every instance whether or not it runs that app.

    CurrentScreenshot: string = ''

    Current screenshot as base64 PNG

    CurrentUrl: string = ''

    Current URL

    Diagnostics?: string

    Compact browser-diagnostics digest for the current step — console errors, failed requests, crashes. Lets the judge explain an infrastructure state (a blank page = ChunkLoadError) instead of hallucinating a reason.

    Goal: string = ''

    The natural-language goal being evaluated

    MaxSteps: number = 0

    Maximum steps for this run

    ScreenshotHistory: string[] = []

    Recent screenshots for progression analysis

    Signal?: AbortSignal

    Cancellation signal for the judge LLM call — see ControllerPromptRequest.Signal. Threaded into AIPromptParams.cancellationToken by Layer 2; not template data.

    StepNumber: number = 0

    Current step number

    StepSummary: string = ''

    Summary of actions taken so far

    ValidationCriteria?: string[]

    Validation-criteria rubric for the run. When present, the judge is asked for a per-criterion {criterion, met, evidence} verdict and Done is derived as all-criteria-met.