openclaw - 💡(How to fix) Fix Dream diary narrative generation times out every night — DREAMS.md not updating [1 comments, 2 participants]

Bug

The dream diary narrative generation (generateAndAppendDreamNarrative) consistently times out, causing DREAMS.md to stop receiving new entries. The dreaming phases themselves (light/rem/deep markdown files) continue to be written correctly since those are direct file operations, but the narrative subagent that produces the diary entries fails every time.

Evidence

Every night since ~April 25, all three phases (light, rem, deep) log timeouts:

memory-core: narrative generation ended with status=timeout for light phase.
memory-core: narrative generation ended with status=timeout for rem phase.
memory-core: narrative generation ended with status=timeout for deep phase.
memory-core: narrative session cleanup failed for light phase: missing scope: operator.admin

Before this date, DREAMS.md was receiving daily narrative entries. After, no new entries appear.

Root Cause

NARRATIVE_TIMEOUT_MS is hardcoded at 15 seconds (15e3) in dreaming-narrative-*.js:

const NARRATIVE_TIMEOUT_MS = 15e3;
// ...
const result = await params.subagent.waitForRun({
    runId,
    timeoutMs: NARRATIVE_TIMEOUT_MS
});
if (result.status !== "ok") {
    params.logger.warn(`memory-core: narrative generation ended with status=${result.status} for ${params.data.phase} phase.`);
    return;  // <-- silently returns, no DREAMS.md write
}

The subagent needs to: start a session, load context, run model inference, and return — all within 15 seconds. For cloud-hosted models with ~600ms+ TTFT alone, this is insufficient. The result is that every narrative generation times out and appendNarrativeEntry() is never called.

Impact

• DREAMS.md receives no new diary entries
• No fallback is attempted — the function returns silently on timeout
• The missing scope: operator.admin cleanup errors suggest the subagent session is orphaned after timeout

Suggested Fix

1. Make NARRATIVE_TIMEOUT_MS configurable (e.g., plugins.entries.memory-core.config.dreaming.narrativeTimeoutMs) or increase the default to at least 60 seconds
2. Consider a deterministic fallback: if the subagent times out, write a structured summary from the phase data directly instead of skipping the diary entry entirely
3. The operator.admin scope error on cleanup after timeout should be handled gracefully

Related

• #67363 — raw verbatim promotion to MEMORY.md (related dreaming output quality issue)
• #67362 — stale cron capture (fixed in 355c92d69b38)

Environment

• OpenClaw 2026.4.26
• Cloud-hosted model (OLLAMA provider with remote inference)