openclaw - 💡(How to fix) Fix bug: subagent announce direct-send error silently swallowed when queue fallback recovers [1 participants]

Problem

When sendSubagentAnnounceDirectly fails but the queue fallback succeeds in runSubagentAnnounceDispatch, the direct-send error is silently swallowed — never logged.

Code path

src/agents/subagent-announce-delivery.ts — sendSubagentAnnounceDirectly catches errors and returns them in the result without logging:

// ~line 512
} catch (err) {
    return {
      delivered: false,
      path: "direct",
      error: summarizeDeliveryError(err),
    };
  }

src/agents/subagent-announce-dispatch.ts — runSubagentAnnounceDispatch records the direct failure as a phase, but when queue fallback succeeds it returns the queue result:

const primaryDirect = await params.direct();
appendPhase("direct-primary", primaryDirect);  // error recorded here...
if (primaryDirect.delivered) {
    return withPhases(primaryDirect);
}
// ...
const fallbackQueue = mapQueueOutcomeToDeliveryResult(fallbackQueueOutcome);
appendPhase("queue-fallback", fallbackQueue);
if (fallbackQueue.delivered) {
    return withPhases(fallbackQueue);  // ...but top-level result is "queued", delivered=true
}

src/agents/subagent-announce.ts — the caller only logs when overall delivery failed:

// ~line 621
if (!delivery.delivered && delivery.path === "direct" && delivery.error) {
    defaultRuntime.error?.(...);
}

When queue recovers: delivered=true, path="queued" → condition is false → error lost.

Observation

The phases array does capture the direct-send failure, but nothing inspects it for logging. The diagnostic information exists in the return value but is never surfaced.

Impact

• Cron jobs with expectsCompletionMessage: true that hit transient direct-send failures (after retry exhaustion) silently fall back to queue delivery with no diagnostic trace.
• Operators cannot distinguish "direct send worked" from "direct send failed, queue recovered" in logs.

Suggested fix

Log the direct-send error at warn level when falling back to queue — either:

1. Inside runSubagentAnnounceDispatch when direct fails before attempting queue fallback
2. In the caller by inspecting delivery.phases for failed direct-primary phases

Option 1 is simpler and keeps the logging close to the decision point.