openclaw - 💡(How to fix) Fix [Feature]: Allow machine-only memory-core dreaming without Dream Diary prose

Summary

Add a supported memory-core dreaming configuration option to disable human-readable Dream Diary prose while keeping the machine-readable dreaming phases and promotion artifacts enabled.

One possible shape:

{
  "plugins": {
    "entries": {
      "memory-core": {
        "config": {
          "dreaming": {
            "enabled": true,
            "humanReadable": {
              "enabled": false
            }
          }
        }
      }
    }
  }
}

Problem to solve

Today, dreaming.enabled=true is useful for the structured memory pipeline, but it also ties that pipeline to human-readable Dream Diary output in DREAMS.md.

Some self-hosted/headless deployments want the light/REM/deep phase artifacts, promotion decisions, and machine-readable memory outputs, but do not want generated diary prose written into the workspace. This matters when the deployment treats memory output as an automation artifact rather than a human journal.

The current workarounds are awkward:

• Disable dreaming entirely, which also loses the machine artifacts.
• Delete or ignore DREAMS.md after the fact, which is reactive and still pays the cost of narrative generation.
• Customize the Dream Diary prompt/language, which helps style and localization but does not express "do not generate human diary prose at all."

Proposed solution

Introduce a backward-compatible sub-option under memory-core dreaming, for example:

"dreaming": {
  "enabled": true,
  "humanReadable": {
    "enabled": false
  }
}

Expected behavior:

• dreaming.enabled=false keeps its existing behavior and disables managed dreaming.
• dreaming.enabled=true with no humanReadable config keeps the current default behavior.
• dreaming.enabled=true and humanReadable.enabled=false still runs the machine dreaming pipeline and writes phase/report artifacts.
• With humanReadable.enabled=false, the Dream Diary narrative subagent is not launched and managed dreaming does not create or append DREAMS.md.

This could also be named differently if the project prefers another config vocabulary, such as dreamDiary.enabled, narrative.enabled, or outputs.humanReadable.

Alternatives considered

• Disable dreaming entirely: not equivalent, because it also disables the useful machine artifacts and promotion flow.
• Ignore or delete DREAMS.md: avoids displaying the prose, but does not prevent generation, writes, indexing noise, or cleanup work.
• Make the Dream Diary prompt configurable: useful and related, but it solves customization rather than the separate need for machine-only dreaming.

Impact

Affected users are operators of self-hosted or automation-first OpenClaw deployments that use memory-core dreaming as infrastructure rather than as a human journal.

Severity is medium for those deployments: the system still runs, but every dreaming sweep can produce unwanted human-readable prose, extra filesystem churn, review noise, and a weaker separation between machine artifacts and human-facing documentation.

The change should be low-risk if the default remains enabled, because existing users keep the current Dream Diary behavior unless they opt out.

Evidence/examples

Related existing issues appear to cover adjacent Dream Diary behavior but not this specific opt-out:

• #79223 asks for configurable Dream Diary language/prompt.
• #79500, #74639, #72546, and #77094 discuss Dream Diary write/report behavior.

This request is narrower: allow deployments to keep managed dreaming enabled while opting out of human-readable Dream Diary prose entirely.

Implementation-wise, this seems testable by asserting that when the option is disabled:

• the managed phase/report artifact path still runs,
• the narrative subagent path is skipped,
• no DREAMS.md write is attempted by managed dreaming,
• default configuration remains unchanged.

Additional information

I am filing this as a feature request rather than a bug because the current behavior appears internally consistent; the missing piece is a supported configuration boundary between machine-readable dreaming artifacts and human-readable diary output.