openclaw-memoryos-memory-admin

Inspect, back up, search, export, and update OpenClaw long-term memory stored with MemoryOS. Use when Codex needs to manage MemoryOS memory files for an Open...

Security Scan

The skill's files, instructions, and required inputs are coherent with its stated purpose of administering MemoryOS long-term memory files; it performs local JSON read/write operations and does not request unrelated credentials or network access.

Purpose & Capability

The name/description match the included CLI script and documentation: the skill only needs a filesystem data root and user/assistant IDs to inspect, back up, export, and mutate MemoryOS JSON files. There are no extraneous env vars, binaries, or unrelated capabilities requested.

Instruction Scope

SKILL.md directs the agent to locate the MemoryOS data root and run the bundled script for targeted operations (summary, search, backup, add, set-profile, export). The instructions reference only MemoryOS file paths and the bundled layout reference; they do not instruct reading unrelated system files or sending data to external endpoints.

Install Mechanism

No install spec is provided (instruction-only with a bundled Python script). No downloads, third-party package installs, or archive extraction are required by the skill package itself.

Credentials

The skill declares no required environment variables, credentials, or config paths. The Python script performs local filesystem operations on a user-supplied data_root and optional backup directory — this is proportional to its purpose. Note: the script will write/overwrite JSON files under the provided paths.

Persistence & Privilege

The registry flags are default (always:false). The agents/openai.yaml sets allow_implicit_invocation: true which permits implicit invocation by the platform — this increases the chance the model will call the skill autonomously, but there are no other elevated privileges requested. This is worth noting but not a disqualifying risk by itself.

Assessment

This skill appears to be what it says: a local MemoryOS JSON admin tool. Before installing or using it, confirm the data_root you pass to the script points to the intended MemoryOS directory (to avoid accidental edits), and always run the provided backup command first. Review the bundled script if you can — it will create/overwrite long_term_user.json and long_term_assistant.json entries, set timestamps to 'now' for new entries, and deliberately writes empty 'knowledge_embedding' arrays for added items (which can temporarily degrade semantic search until embeddings are rebuilt by your runtime). If you want to limit autonomous changes, note that the skill metadata allows implicit invocation by the agent — consider disabling implicit/autonomous invocation for agents you don't trust or explicitly invoking the skill only when needed.

Manage OpenClaw long-term memory by operating directly on MemoryOS storage files and by following the MemoryOS data model used in long_term_user.json and long_term_assistant.json.

Use the bundled script for deterministic operations. Read the reference file only when you need to confirm the on-disk layout or explain how MemoryOS maps user profile, user knowledge, and assistant knowledge.

Quick Start

Prefer scripts/memoryos_admin.py over ad hoc edits.

Common commands:

python scripts/memoryos_admin.py summary --data-root "D:\path\to\memoryos_data" --user-id alice --assistant-id openclaw
python scripts/memoryos_admin.py search-user --data-root "D:\path\to\memoryos_data" --user-id alice --query "project preference"
python scripts/memoryos_admin.py add-user-knowledge --data-root "D:\path\to\memoryos_data" --user-id alice --text "User prefers concise technical answers."
python scripts/memoryos_admin.py set-profile --data-root "D:\path\to\memoryos_data" --user-id alice --profile-file ".\profile.txt"
python scripts/memoryos_admin.py backup --data-root "D:\path\to\memoryos_data" --user-id alice --assistant-id openclaw

Workflow

Follow this order unless the user asks for a narrower task:

1. Locate the MemoryOS data root used by the OpenClaw integration.
2. Run summary to inspect current profile and knowledge counts.
3. Run backup before any mutating action.
4. Use search-user or search-assistant to verify whether the target memory already exists.
5. Use add-user-knowledge, add-assistant-knowledge, or set-profile for intentional updates.
6. Use export-markdown when the user wants a readable audit or handoff artifact.

Data Rules

Treat these files as the authoritative long-term memory state:

• users/<user_id>/long_term_user.json
• assistants/<assistant_id>/long_term_assistant.json

Interpret them this way:

• user_profiles.<user_id>.data: current synthesized user profile
• knowledge_base[]: user-specific long-term knowledge entries
• assistant_knowledge[]: assistant-side reusable knowledge entries

Preserve existing timestamps unless you are explicitly adding or replacing content.

Do not hand-edit embeddings. The bundled admin script stores and updates records without embeddings on purpose, because this skill is for memory administration, curation, backup, and inspection. Semantic retrieval inside live MemoryOS runtime can rebuild or add embeddings through normal MemoryOS flows when needed.

Mutation Guidance

Use set-profile only when the user wants to replace the synthesized profile with a curated version.

Use add-user-knowledge for durable facts about the user, preferences, identity, recurring tasks, or persistent context that OpenClaw should remember.

Use add-assistant-knowledge for workflow instructions, project background, house rules, or reusable operational knowledge that belongs to the assistant rather than the user.

Before appending a near-duplicate memory, search first and ask whether the old entry should be kept, replaced manually, or left unchanged if the instruction is ambiguous.

OpenClaw Integration Notes

If OpenClaw is not present in the workspace, treat OpenClaw as the consuming agent and MemoryOS as the storage backend. The critical integration contract is the MemoryOS data_storage_path and the chosen user_id plus assistant_id.

When wiring another agent project to MemoryOS:

• keep a stable user_id per person
• keep a stable assistant_id per OpenClaw persona or deployment
• point the agent to the same data_storage_path
• back up memory files before schema or workflow changes

References

Read references/memoryos-layout.md when you need exact file layout and field names.