Member Junction
    Preparing search index...

    The accumulated outcome of one harness turn, assembled by HarnessAgentBase from the adapter's event stream.

    interface HarnessTurnResult {
        ArtifactUrls?: string[];
        AsyncWakePending?: boolean;
        CloudExecutionState?: CloudExecutionState;
        CostUsd?: number;
        ErrorMessage?: string;
        InputTokens: number;
        OutputTokens: number;
        PullRequestUrl?: string;
        RawText: string;
        ReportedModel?: string;
        SessionId?: string;
        ThinkingTokens?: number;
    }
    Index

    Properties

    ArtifactUrls?: string[]

    Optional artifact URLs reported by a cloud agent run.

    AsyncWakePending?: boolean

    True when the turn detached cleanly in adaptive-wake mode and awaits the next AgentWake.

    CloudExecutionState?: CloudExecutionState

    Terminal or detached cloud execution state when driven by a BaseCloudHarnessAdapter.

    CostUsd?: number
    ErrorMessage?: string

    Set when the turn failed rather than completed.

    InputTokens: number

    Summed usage for the turn. Zeros when the harness reports none — which is itself a finding.

    OutputTokens: number
    PullRequestUrl?: string

    Optional pull request URL created or updated by a cloud agent run.

    RawText: string

    Raw turn-end text, handed to the Loop JSON parser exactly as a prompt response would be.

    ReportedModel?: string

    The model the harness ACTUALLY used, as it reported it (e.g. claude-opus-4-6).

    Distinct from the model we asked for. A harness free to pick its own model will, and recording the one we assumed instead of the one it used makes cost attribution wrong — Opus and Sonnet are not the same price.

    SessionId?: string

    Vendor session id, persisted to AIAgentRun.ExternalSessionID for resume and log correlation.

    ThinkingTokens?: number