Self Learning Skill

The stated self-learning purpose matches the core behavior: reading conversation history, analyzing memory/config files, and updating learning records. The concern is proportionality: the prompt explicitly treats API keys, tokens, channel IDs, and similar configuration secrets as memory-worthy content, which can turn transient sensitive data into durable plaintext records.

The documentation describes automatic updates and deletion protections, but the runtime path defaults to direct writes unless --dry-run is manually used, and the code does not enforce the documented delete confirmation or maximum-delete controls before applying AI-generated operations.

Installation is conventional: clone the package, install Python requirements, and optionally copy an OpenClaw hook. The hook behavior is documented, but enabling it allows session-start prompt injection of reminders and prompt-submit monitoring for correction signals.

The skill reads recent OpenClaw session history, current memory/config files, and /root/.openclaw/openclaw.json for workspace discovery. That is related to the purpose, but it is broad and under-scoped for a memory tool because it may collect multiple sessions and infer a workspace without an explicit path.

The skill creates persistent .learnings, logs, daily memory, backups, execution history, and overwrites core workspace markdown files. Promotion can also write caller-selected target files inside or potentially outside the workspace, so persistence authority is broader than the documented core memory files.