openclaw - 💡(How to fix) Fix Dreaming needs supported guard to disable session transcript ingestion during QMD migrations

Official PRs (…)
ON THIS PAGE

Recommended Tools

×6

Utilities matched from this issue’s tags and category — try them while you read without losing context.

GitHub issue graph ai analysis

Paste a GitHub issue URL. We fetch that issue, discover linked issues from bodies/comments/timeline, collect linked pull requests, and produce a structured English report.

The report is written in English Markdown for sharing and archival.

Helpful · Quick feedback

Loading…

During staged QMD memory migrations, session transcript indexing into dreaming/QMD may intentionally be out of scope. In that state, operators need a supported way to disable session transcript ingestion so volatile session, cron, and tool-output fragments do not enter dreaming memory.

Observed across the 2026.5.26 to 2026.5.27 update window. Current installed version observed locally:

OpenClaw 2026.5.27 (27ae826)

A local environment used sentinel files named:

memory/.dreams/disable-session-ingestion

and a local quality filter to keep volatile session/cron/tool output from entering dreaming memory. Package/runtime updates can overwrite that local installed-runtime guard. When the guard was lost, non-main workspaces regenerated memory/.dreams/session-corpus files and volatile dream stores accumulated raw cron/session snippets.

Root Cause

During staged QMD memory migrations, session transcript indexing into dreaming/QMD may intentionally be out of scope. In that state, operators need a supported way to disable session transcript ingestion so volatile session, cron, and tool-output fragments do not enter dreaming memory.

Observed across the 2026.5.26 to 2026.5.27 update window. Current installed version observed locally:

OpenClaw 2026.5.27 (27ae826)

A local environment used sentinel files named:

memory/.dreams/disable-session-ingestion

and a local quality filter to keep volatile session/cron/tool output from entering dreaming memory. Package/runtime updates can overwrite that local installed-runtime guard. When the guard was lost, non-main workspaces regenerated memory/.dreams/session-corpus files and volatile dream stores accumulated raw cron/session snippets.

Fix Action

Fix / Workaround

Local workaround

The local installed dreaming runtime was patched to:

Code Example

OpenClaw 2026.5.27 (27ae826)

---

memory/.dreams/disable-session-ingestion
RAW_BUFFERClick to expand / collapse

Summary

During staged QMD memory migrations, session transcript indexing into dreaming/QMD may intentionally be out of scope. In that state, operators need a supported way to disable session transcript ingestion so volatile session, cron, and tool-output fragments do not enter dreaming memory.

Observed across the 2026.5.26 to 2026.5.27 update window. Current installed version observed locally:

OpenClaw 2026.5.27 (27ae826)

A local environment used sentinel files named:

memory/.dreams/disable-session-ingestion

and a local quality filter to keep volatile session/cron/tool output from entering dreaming memory. Package/runtime updates can overwrite that local installed-runtime guard. When the guard was lost, non-main workspaces regenerated memory/.dreams/session-corpus files and volatile dream stores accumulated raw cron/session snippets.

Expected behavior

OpenClaw should provide a supported configuration switch or durable sentinel mechanism to disable session transcript ingestion into dreaming/QMD during staged migrations.

When disabled:

  • Dreaming should not create active memory/.dreams/session-corpus files.
  • Dreaming should not promote raw command/status/transcript snippets into short-term recall or phase-signal stores.
  • The disable setting should survive package updates.

Actual behavior

Without the local installed-runtime guard, a Memory Dreaming Promotion run recreated active session-corpus files and volatile dream entries from raw session/cron fragments.

The fallout was not limited to the main agent. Non-main workspaces such as cron-related agents also had session-corpus material that required cleanup.

Local workaround

The local installed dreaming runtime was patched to:

  • check for memory/.dreams/disable-session-ingestion;
  • skip session transcript ingestion when present;
  • apply a local blocklist/quality filter for volatile command/status snippets.

Operational cleanup after the guard was lost:

  • quarantined generated session-corpus files;
  • scrubbed session-corpus entries from active volatile dream stores;
  • forced QMD reindex for affected agents;
  • recreated sentinel files for pilot agents.

Verification evidence

After restoring the local guard:

  • pilot agents had memory/.dreams/disable-session-ingestion;
  • active session-corpus file count was 0;
  • QMD memory status was clean for pilot agents;
  • no active volatile session-corpus files reappeared after subsequent monitor checks;
  • gateway health remained OK.

Requested upstream outcome

Please add a first-class setting or supported sentinel for disabling session transcript ingestion into dreaming/QMD, plus tests ensuring package updates do not remove the behavior. This would let staged memory migrations keep QMD clean until transcript indexing is explicitly enabled.

Vote matrix · Quick signals

Works
Did the solution work? Tap to confirm.
Easy Fix
Was it a quick fix?
Time Saver
Did it save you time?
Blocking
Was it severely blocking?
Common Issue
Are others likely hitting this too?
Flaky / Intermittent
Is it intermittent?
Verified / Reproducible
Can you reproduce it reliably?
Loading…

FAQ

Expected behavior

OpenClaw should provide a supported configuration switch or durable sentinel mechanism to disable session transcript ingestion into dreaming/QMD during staged migrations.

When disabled:

  • Dreaming should not create active memory/.dreams/session-corpus files.
  • Dreaming should not promote raw command/status/transcript snippets into short-term recall or phase-signal stores.
  • The disable setting should survive package updates.

Still need to ship something?

×6

Another batch ranked right after the header list — different links, same matching logic.

Back to top recommendations

TRENDING