"sessionStrategy": "memoryReflection" extremely slow and no effective output

[qaqcvc]

Plugin Version

1.1.0-beta.9

OpenClaw Version

2026.3.13

Bug Description

"sessionStrategy": "memoryReflection"
Today I enabled sessionStrategy with memoryReflection, but it takes very long time to finish a session recap (around 8 minutes, time out does not work) after sending /new in openclaw. And in the .learning folder, LEARNINGS.md repeately logging meaningless record.

# Learnings

Append structured entries:
- LRN-YYYYMMDD-XXX for corrections / best practices / knowledge gaps
- Include summary, details, suggested action, metadata, and status


## [LRN-20260319-001] best_practice

**Logged**: 2026-03-19T21:15:39.602Z
**Priority**: medium
**Status**: triage
**Area**: config

### Summary
Investigate last failed tool execution and decide whether it belongs in .learnings/ERRORS.md.

### Details
The reflection pipeline fell back; confirm the failure is reproducible before treating it as a durable error record.

### Suggested Action
Reproduce the latest failed tool execution, classify it as triage or error, and then log it with the appropriate tool/file path evidence.

### Metadata
- Source: memory-lancedb-pro/reflection:memory/reflections/2026-03-19/211349568-main-efb39d2a-eaaf-4d6f-9608-54e61ac0.md
---


## [LRN-20260319-002] best_practice

**Logged**: 2026-03-19T21:17:36.701Z
**Priority**: medium
**Status**: triage
**Area**: config

### Summary
Investigate last failed tool execution and decide whether it belongs in .learnings/ERRORS.md.

### Details
The reflection pipeline fell back; confirm the failure is reproducible before treating it as a durable error record.

### Suggested Action
Reproduce the latest failed tool execution, classify it as triage or error, and then log it with the appropriate tool/file path evidence.

### Metadata
- Source: memory-lancedb-pro/reflection:memory/reflections/2026-03-19/211547146-main-efb39d2a-eaaf-4d6f-9608-54e61ac0.md
---


## [LRN-20260319-003] best_practice

**Logged**: 2026-03-19T21:22:38.022Z
**Priority**: medium
**Status**: triage
**Area**: config

### Summary
Investigate last failed tool execution and decide whether it belongs in .learnings/ERRORS.md.

### Details
The reflection pipeline fell back; confirm the failure is reproducible before treating it as a durable error record.

### Suggested Action
Reproduce the latest failed tool execution, classify it as triage or error, and then log it with the appropriate tool/file path evidence.

### Metadata
- Source: memory-lancedb-pro/reflection:memory/reflections/2026-03-19/212035646-main-ffd31194-f270-4f72-8e5b-00284061.md
---


## [LRN-20260319-004] best_practice

**Logged**: 2026-03-19T21:24:38.563Z
**Priority**: medium
**Status**: triage
**Area**: config

### Summary
Investigate last failed tool execution and decide whether it belongs in .learnings/ERRORS.md.

### Details
The reflection pipeline fell back; confirm the failure is reproducible before treating it as a durable error record.

### Suggested Action
Reproduce the latest failed tool execution, classify it as triage or error, and then log it with the appropriate tool/file path evidence.

### Metadata
- Source: memory-lancedb-pro/reflection:memory/reflections/2026-03-19/212248026-main-ffd31194-f270-4f72-8e5b-00284061.md
---

      "memory-lancedb-pro": {
        "enabled": true,
        "config": {
          "embedding": {
            "apiKey": "xxx",
            "model": "jina-embeddings-v5-text-small",
            "baseURL": "https://api.jina.ai/v1",
            "dimensions": 1024,
            "taskQuery": "retrieval.query",
            "taskPassage": "retrieval.passage",
            "normalized": true
          },
          "autoRecall": true,
          "autoRecallMinLength": 8,
          "retrieval": {
            "mode": "hybrid",
            "vectorWeight": 0.7,
            "bm25Weight": 0.3,
            "minScore": 0.45,
            "hardMinScore": 0.55,
            "candidatePoolSize": 20,
            "rerank": "cross-encoder",
            "rerankProvider": "jina",
            "rerankApiKey": "xxx",
            "rerankModel": "jina-reranker-v3",
            "rerankEndpoint": "https://api.jina.ai/v1/rerank",
            "recencyHalfLifeDays": 14,
            "recencyWeight": 0.1,
            "timeDecayHalfLifeDays": 60,
            "filterNoise": true,
            "lengthNormAnchor": 500
          },
          "autoCapture": true,
          "selfImprovement": {
            "enabled": true,
            "beforeResetNote": true,
            "ensureLearningFiles": true
          },
          "mdMirror": {
            "enabled": true,
            "dir": "/root/.openclaw/workspace/memory/md-mirror"
          },
          "llm": {
            "apiKey": "xxx",
            "model": "gpt-4o-mini",
            "baseURL": "https://api.openai.com/v1"
          },
          "sessionStrategy": "memoryReflection",
          "memoryReflection": {
            "storeToLanceDB": true,
            "injectMode": "inheritance+derived",
            "messageCount": 80,
            "maxInputChars": 20000,
            "thinkLevel": "low",
            "timeoutMs": 30000
          }
        }
      }
    }
  }
}

Expected Behavior

previsouly without enabling this memoryReflection, I also received the message of

/note self-improvement (before reset):

• If anything was learned/corrected, log it now:
• .learnings/LEARNINGS.md (corrections/best practices)
• .learnings/ERRORS.md (failures/root causes)
• Distill reusable rules to AGENTS.md / SOUL.md / TOOLS.md.
• If reusable across tasks, extract a new skill from the learning.
• Then proceed with the new session.

Surely nothing works though.

Steps to Reproduce

...

Error Logs / Screenshots

...

Embedding Provider

Jina

OS / Platform

Ubuntu

[AliceLJY]

Hi @qaqcvc, thanks for reporting this.

From your LEARNINGS.md output, I can see the reflection pipeline is consistently falling back to a hardcoded template — the repeated "Investigate last failed tool execution" entries are not real learnings, they're the fallback text being incorrectly treated as governance candidates. That part is a bug on our side and we'll fix it (fallback text should not produce LEARNINGS.md entries).

To diagnose the slowness and root cause (why both the embedded and CLI reflection runners are failing), could you provide:

1. Gateway logs during a /new command — look for lines containing memory-reflection:, they will show which runner was attempted and why it failed
2. Your agent model config — the reflection embedded runner uses your agent's primary model (not the llm.model in the plugin config). Can you confirm what model is configured for the agent itself?

In the meantime, if you want to stop the repetitive LEARNINGS.md entries, you can set "selfImprovement": { "enabled": false } until we ship the fix.