LLM Memory Integration

This package appears coherent for a local LLM+vector memory integration, but it will read and write your local memory database (~/.openclaw/memory-tdai/vectors.db), persona.md, caches and logs, and will send text to whatever embedding/LLM endpoints you configure. Things to check before installing: - Verify and supply only trusted embedding/LLM endpoints and keys; any text sent there may be stored or logged by the provider. - Inspect or vet the SQLite vector extension file (~/.openclaw/extensions/...) before allowing the skill to load it — a native .so can execute arbitrary code at load time. The project itself warns to verify the extension source. - Review the persona and vectors DB for sensitive data; the skill will update persona.md and write vectors into vectors.db. - The scripts run subprocesses and the sqlite3 CLI to query/load extensions; while implemented using parameterized calls (no shell=True), ensure the filesystem paths used are expected and not symlinked to sensitive locations. - If you need stronger assurance, run the scripts in an isolated environment (container or VM) and examine the full omitted files (35 files truncated here) before granting access to your production workspace. I have medium confidence in this assessment because the code and SKILL.md are internally consistent, but the presence of native extension loading and broad read/write to a local memory DB are sensitive and merit user review.