Memory Setup (OpenClaw)

Configure and validate OpenClaw memory recall for persistent context. Use when enabling memory_search/memory_get, fixing poor memory recall, or setting up ME...

Security Scan

The instructions appear to match a memory-setup purpose, but there are multiple inconsistencies (undeclared CLI and API-key requirements, and metadata mismatches) that make the package suspicious until clarified.

Purpose & Capability

The SKILL.md clearly targets configuring OpenClaw memory (workspace files, config keys, providers). However the skill metadata declares no required binaries or environment variables even though the instructions assume an 'openclaw' CLI and may require remote provider API keys (OpenAI/Gemini/etc.). The registry ownerId/slug in the provided metadata also doesn't match the _meta.json ownerId/slug, which is an additional coherence problem.

Instruction Scope

Instructions themselves are narrowly focused on configuring memory files, updating agents.defaults.memorySearch, restarting the gateway, and testing memory_search/memory_get — all appropriate for the stated purpose. They do not instruct reading arbitrary unrelated files or exfiltrating data. However they assume access to configuration and CLI control (restart gateway, run 'openclaw status' / 'openclaw memory status') that should have been reflected in the declared requirements.

Install Mechanism

This is an instruction-only skill with no install spec and no code files, so there is no download/execution install risk. Nothing is written to disk by the skill package itself.

Credentials

SKILL.md explicitly references using remote memory providers and setting their API keys (env var or memorySearch.remote.apiKey), but requires.env and primary credential fields are empty. That mismatch means the skill may need secrets (service API keys) not declared in metadata — the user would not be warned up front about credential access.

Persistence & Privilege

always is false and there is no install-time persistence requested. The skill does not claim to modify other skills or system-wide settings beyond advising edits to OpenClaw config files (which is within its stated purpose).

What to consider before installing

This skill's instructions look plausible for configuring OpenClaw memory, but there are several inconsistencies you should resolve before installing or following them: - Verify the author/owner: the registry metadata ownerId and the _meta.json ownerId differ. Ask the publisher which identity is correct. - Confirm prerequisites: the SKILL.md assumes you have an 'openclaw' CLI and the ability to restart the gateway. The package metadata does not declare these required binaries — make sure you actually have the CLI and appropriate permissions. - API keys: the doc says remote providers require API keys, but no env vars are declared. Only set API keys with least privilege and scope they need, and prefer storing them in your agent's secure secrets rather than pasting them into workspace files. - Backup config: before modifying agents.defaults.memorySearch or restarting services, back up your OpenClaw config and any policies. - Principle of least privilege: if you must provide remote model API keys, consider using keys scoped to a single project/account and monitor their use. If the publisher can explain the metadata mismatches and update the skill metadata to list required binaries and env vars (or sign the package to a verified owner), this would reduce the suspicion.

Set up durable memory for OpenClaw so the agent can recall prior decisions, preferences, and todos.

1) Prepare workspace files

In workspace root, keep:

• MEMORY.md (curated long-term memory)
• memory/YYYY-MM-DD.md (daily notes)

Optional structure:

• memory/projects/
• memory/system/
• memory/groups/

2) Enable memory search in OpenClaw config

Configure under agents.defaults.memorySearch (not top-level memorySearch).

Example:

{
  "agents": {
    "defaults": {
      "memorySearch": {
        "enabled": true,
        "provider": "local",
        "includeDefaultMemory": true,
        "maxResults": 20,
        "minScore": 0.3
      }
    }
  }
}

Notes:

• includeDefaultMemory: true indexes MEMORY.md + memory/**/*.md.
• Providers: local, openai, gemini, voyage, mistral.
• If using remote provider, set corresponding API key (env var or memorySearch.remote.apiKey).

3) Restart and verify

• Restart gateway after config changes.
• Verify with:
  • openclaw status
  • openclaw memory status (if available in your CLI build)

4) Test recall behavior

Ask a question about past context, then verify the agent:

1. Runs memory_search
2. Uses memory_get for precise lines when needed
3. Cites source path/lines when useful

5) Troubleshooting

memory_search unavailable

• Ensure agents.defaults.memorySearch.enabled = true.
• Ensure policy allows memory tools.
• Restart gateway.

low-quality matches

• Lower minScore (e.g. 0.2) to broaden hits.
• Raise maxResults (e.g. 30).
• Write more specific notes in MEMORY.md and daily logs.

local provider issues

• Confirm local model path/settings if configured.
• If needed, switch provider to remote and set API key.

6) Recommended operating rule

Before answering questions about prior work, decisions, dates, people, preferences, or todos:

1. memory_search first
2. memory_get second (only needed lines)
3. Say you checked memory if confidence is still low

This keeps responses grounded and auditable.