Bug type

Regression (worked before, now fails)

Beta release blocker

No

Summary

With dreaming enabled, OpenClaw creates many internal dreaming-narrative-{light|rem|deep}-* sessions overnight and surfaces them in the normal Sessions list, even though the managed dreaming cron job is configured for a daily 0 3 * * * sweep.

Steps to reproduce

1. Run OpenClaw 2026.4.11 on macOS with plugins.entries.memory-core.config.dreaming.enabled = true.
2. Leave OpenClaw running overnight.
3. Open the Sessions list the next morning.
4. Observe many sessions named dreaming-narrative-light-*, dreaming-narrative-rem-*, and dreaming-narrative-deep-*.
5. Confirm in ~/.openclaw/cron/jobs.json that the managed dreaming job is scheduled as 0 3 * * *.
6. Confirm in gateway logs that dreaming phases actually ran repeatedly between 03:00 and 08:31 on April 12, 2026.

Expected behavior

With the managed dreaming cron job scheduled as 0 3 * * *, dreaming should run as a single scheduled sweep per day rather than repeatedly creating many internal narrative sessions over several hours. Internal dreaming diary sessions should also not flood the normal Sessions list.

Actual behavior

The Sessions list filled up overnight with internal sessions such as dreaming-narrative-light-*, dreaming-narrative-rem-*, and dreaming-narrative-deep-*. Local evidence shows 30 dream-diary session transcripts were created: 14 light, 14 rem, and 2 deep. Gateway logs show repeated dreaming activity between 03:00 and 08:31 on April 12, 2026, instead of a single daily run.

OpenClaw version

2026.4.11 (769908e)

Operating system

macOS 26.4 / Darwin 25.4.0 (arm64)

Install method

npm global

Model

openai-codex/gpt-5.4-mini

Provider / routing chain

openclaw -> openai-codex

Additional provider/model setup details

• plugins.entries.memory-core.config.dreaming.enabled = true
• No dreaming frequency override was found in ~/.openclaw/openclaw.json
• Managed cron job in ~/.openclaw/cron/jobs.json:
  • name: Memory Dreaming Promotion
  • schedule: 0 3 * * *
  • wakeMode: next-heartbeat

Logs, screenshots, and evidence

Picture attached

Impact and severity

Severity: Annoying
Frequncy: Night/Always
Consequence: High token usage

Additional information

• Dream diary docs state that memory-core runs a best-effort background subagent turn for diary entries, which explains why separate dreaming-narrative-* sessions are created.

Logs, screenshots, and evidence

• Screenshot of the Sessions picker showing many dreaming-narrative-* sessions
• ~/.openclaw/openclaw.json shows:
  • plugins.entries.memory-core.config.dreaming.enabled = true
• ~/.openclaw/cron/jobs.json shows managed job:
  • Memory Dreaming Promotion
  • cron expression: 0 3 * * *
• ~/.openclaw/logs/gateway.log shows repeated dreaming runs on 2026-04-12, including:
  • 03:00:01 light dreaming staged
  • 03:12:44 light dreaming staged
  • 03:20:00 light dreaming staged
  • 04:20:01 light dreaming staged
  • 04:30:49 light dreaming staged
  • 05:12:44 light dreaming staged
  • 05:20:01 light dreaming staged
  • 06:20:01 light dreaming staged
  • 06:30:49 light dreaming staged
  • 07:12:43 light dreaming staged
  • 07:20:01 light dreaming staged
  • 08:20:00 light dreaming staged
  • 08:30:49 light dreaming staged
• The same log also shows repeated dream diary writes for light/rem, and some deep-phase diary writes.
• Session transcript directory contains 30 matching dream-diary session files:
  • 14 light
  • 14 rem
  • 2 deep

Impact and severity
This clutters the normal Sessions UI with internal background sessions and makes it harder to find user-relevant conversations. It also indicates that dreaming is executing far more often than the configured daily cron cadence, which may increase token usage and create confusing background activity. In the observed case, this produced 30 internal dream-diary sessions overnight.