Convert Memory Files Between Systems
ReviewAudited by ClawScan on May 10, 2026.
Overview
The migration goal is understandable, but the included scripts directly modify OpenClaw memory and plugin configuration with hard-coded paths and no clear backup, review, or rollback controls.
Use this only if you understand and want the memory migration. Before running any script, replace hard-coded paths with your own, back up the OpenClaw config and memos database, review the markdown memories for sensitive or instruction-like content, and test with a dry run or copy of the database first.
Findings (4)
Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.
Running this as-is could alter or duplicate persistent memory records and may affect how the agent behaves in later sessions.
The helper directly writes every matching markdown memory file from a hard-coded local directory into a persistent SQLite memory database, with no backup, dry run, duplicate handling, or user-selected path controls.
conn = sqlite3.connect('/home/hahaha1234/.openclaw/memos-local/memos.db') ... for filename in os.listdir(memory_dir) ... INSERT INTO chunksRequire explicit user confirmation, parameterize the source and database paths, create a backup first, and add dry-run plus deduplication checks before writing to the database.
A mistaken run could disable or break a memory plugin configuration and change the agent's runtime behavior.
The helper persistently changes OpenClaw plugin configuration by removing a plugin from entries, load paths, and allow lists, but it does not create a backup or provide rollback safeguards.
del config['plugins']['entries']['memory-lancedb-pro'] ... config['plugins']['allow'].remove('memory-lancedb-pro') ... json.dump(config, file, indent=2)Back up the configuration file, validate the config structure, show the planned diff to the user, and only save after explicit approval.
Old or untrusted memory text, including instruction-like content, could be reused by the agent in future tasks and influence later responses.
Raw markdown memory content is inserted into the memos-local persistent memory table without review, filtering, or trust-boundary handling.
content = file.read() ... INSERT INTO chunks ... VALUES (?, ?, ?, ?, ?, ?, ?)
Review and filter imported memories before insertion, avoid importing instruction-like or untrusted content blindly, and provide a way to audit and remove imported records.
The mismatch can confuse users or reviewers about which script should be run for a configuration-changing operation.
The documentation references a delete_memory_lancedb_pro_config.py companion script, but the provided manifest contains remove_memory_lancedb_pro.py instead.
scripts/delete_memory_lancedb_pro_config.py — 删除配置文件中 memory-lancedb-pro 相关内容的 Python 脚本 ... scripts/remove_memory_lancedb_pro.py — automation script
Make the companion script names consistent and ensure the documented file is exactly the one included and reviewed.