ART Framework API Docs
    Preparing search index...

    Interface HITLFeedback

    The standardized feedback structure returned by the user.

    This is the canonical format for user feedback. The framework converts this into a successful tool result when resuming execution. The LLM sees this as a normal tool completion.

    interface HITLFeedback {
        approved: boolean;
        metadata?: Record<string, unknown>;
        modifiedArgs?: Record<string, unknown>;
        reason?: string;
        selectedValues?: (string | number | boolean)[];
        textInput?: string;
        timestamp: number;
        value?: unknown;
    }
    Index

    Properties

    approved: boolean

    Whether the action was approved/confirmed. For non-boolean input types, this is true if the user submitted (vs. cancelled/dismissed).

    metadata?: Record<string, unknown>

    Optional metadata about the feedback context.

    modifiedArgs?: Record<string, unknown>

    Modified tool arguments if allowModifyArgs was true and user modified them.

    reason?: string

    Optional reason/comment provided by the user. Particularly useful for rejections to explain why.

    selectedValues?: (string | number | boolean)[]

    For select/multiselect, the selected value(s).

    Use value instead

    textInput?: string

    For text input type, the raw text entered.

    Use value instead

    timestamp: number

    Timestamp when feedback was provided.

    value?: unknown

    The user's input value, type depends on the inputType:

    • boolean: true/false
    • text: string
    • number: number
    • select: the selected option's value
    • multiselect: array of selected values
    • date/datetime: ISO string
    • file: file reference object
    • confirm: undefined (use approved field)
    • custom: matches customSchema