openclaw - 💡(How to fix) Fix Bug: Dreaming session corpus ingestion stops processing new sessions after 5/15

Bug Report: Dreaming Session Corpus Ingestion Stops Processing New Sessions After 5/15

Summary

The memory-core plugin's session corpus ingestion pipeline stopped processing new session transcripts after 2026-05-15. Sessions created after that date are NOT in seenMessages and are NOT written to session-corpus/, even though the Dreaming cron runs daily and reports status: ok.

Evidence

Session-corpus state:

• Latest corpus file: 2026-05-15.txt (425 bytes) — no files written since 5/15
• session-ingestion.json last modified: 2026-05-15 03:02
• Tracked files: 352 (all pre-5/15)
• seenMessages scopes: 1592

Actual sessions on disk:

• Total session .jsonl files: 466 (177 after 5/15)
• Sessions after 5/15 NOT in seenMessages: 175
• Only 2 sessions after 5/15 were ingested

Dreaming cron:

• Runs daily at 3:00 AM, all status: ok
• Recent run summaries: all NO_REPLY
• No new content written to DREAMS.md since 5/15

What I verified is NOT the cause:

• ✅ memory-core plugin enabled: true, dreaming enabled: true
• ✅ Session files exist at ~/.openclaw/agents/main/sessions/*.jsonl
• ✅ listSessionFilesForAgent(agentId) returns correct paths
• ✅ sessionPathForFile() builds correct keys (main:sessions/main/{uuid}.jsonl)
• ✅ Session ingestion state file is valid JSON, no corruption
• ✅ No gateway errors related to memory-core

OpenClaw version: v2026.5.20 (e510042)

Key Finding

The collectSessionIngestionBatches function lists all session files correctly via listSessionFilesForAgent(), but the ingestion loop appears to skip new sessions entirely. The ingestion state (session-ingestion.json) still references old session files with state keys like main:sessions/main/{uuid}.jsonl, but new sessions created after 5/15 never enter the ingestion pipeline.

The ingestion state seenMessages has 1592 scopes, but 175 new sessions after 5/15 are NOT represented in it. The files dict has 352 entries, all pre-5/15.

Possible Causes

1. Workspace matching issue in resolveMemoryDreamingWorkspaces() — the workspace key comparison may fail if path resolution changed
2. The ingestion lookbackDays or cutoffMs calculation may be excluding newer sessions
3. A change in how buildSessionScopeKey() or buildSessionStateKey() generates keys between versions
4. The ingestion state file may be silently failing to update (changed flag not triggered)

Impact

• No new session content is fed into Dreaming phases (Light/REM/Deep)
• No new content for DREAMS.md diary entries
• Short-term recall still works (8690 entries), but session transcripts are the primary source for new memories

Environment

• OS: Linux 6.6.87.2-microsoft-standard-WSL2 (x64)
• Node: v22.22.2
• OpenClaw: v2026.5.20
• Workspace: /mnt/e/OpenClaw
• memory-core: bundled

Happy to provide additional debug info or test patches.