AutoBE
    Preparing search index...

    Predicate-side failure indicator written by AutoBeReplayComputer summarize when a phase finished cleanly but its post-execution predicate still returned false.

    Closes the "왜 멈췄는지" gap for the cases where the agent loop did not throw and no compile diagnostic surfaced — without this field, the only signal of failure was success: false with reason: null and compile: null.

    kind is grep-friendly and stable across history shape changes; count carries the quantitative payload (number of missed operations, number of operation failures) so a reviewer can tell a single-failure phase apart from a many-failure phase without re-reading the underlying history.

    interface IPhasePredicateFailure {
        count?: number;
        kind: "unknown" | "interface_missed" | "realize_operation_failures";
    }
    Index

    Properties

    Properties

    count?: number

    Quantitative payload paired with the predicate-failure kind. Present for "interface_missed" (missed count) and "realize_operation_failures" (operation-failure count); absent for "unknown" because by definition the summarizer could not derive a meaningful number.

    kind: "unknown" | "interface_missed" | "realize_operation_failures"

    Predicate-failure class:

    • "interface_missed": interface phase finished but IAutoBeInterfaceHistory.missed.length > 0. count carries the number of missed operations.
    • "realize_operation_failures": realize phase compile passed but IAutoBeRealizeHistory.operationFailures.length > 0. count carries the number of failed operations.
    • "unknown": defensive fallback. The summarizer detected success: false without a known cause (no compile failure, no caught reason, no recognized predicate signal). Set so the failure-reason invariant ("success: false ⇒ at least one of reason / compile.kind / predicateFailure.kind is populated") holds even when a future predicate adds a new false-branch the summarizer has not been taught about yet.