Super Memori

This package appears to implement what it claims, but review and accept the privacy implications before installing. Specific recommendations: - Review or disable the auto-learner: it reads ~/.openclaw/logs/commands.log and writes discovered commands into pending JSONs; that can capture sensitive commands (passwords typed into shell, tokens, SSH commands, etc.). If you don't want that, don't enable/cron it or remove/modify auto-learner.sh. - Inspect which directories will be indexed (defaults to ~/.openclaw/workspace/memory) and make sure no files with secrets (keys, credentials, private notes) are present. The embedding scripts read file contents and store vectors + content snippets in the local vector DB. - Be aware SentenceTransformer will attempt to download model weights (e.g., from Hugging Face) if the model isn't cached locally — that causes outbound network traffic and increases the attack surface; to avoid this, pre-install models in an offline environment. - Confirm Qdrant is a local instance (127.0.0.1:6333) before running; the code targets localhost but will fail if misconfigured. Running in an isolated/test host first is prudent. - Check file permissions for the skill's directories (~/.openclaw/...) and the SQLite DB to limit access to only intended users. - If you need stronger guarantees, run the skill in an isolated VM/container, audit the code paths (auto-learner, embed-*), and run health-check.sh and eval-memory.sh in JSON mode to inspect reported state before enabling indexing or scheduling cron jobs. - The package does not request external credentials explicitly, but it will read local files; treat that as a manual decision about what to expose to the skill's memory stores.