MemOS Dreaming

Automatically consolidates daily memories by scoring and filtering entries from MemOS SQLite and daily logs, writing top insights to MEMORY.md each morning.

What to consider before installing/running this skill: - Back up your data first: make manual backups of ~/.openclaw/workspace/MEMORY.md and ~/.openclaw/memos-local/memos.db before running this skill or audit with --apply. The scripts will write to MEMORY.md and the audit's --apply mode can remove/modify entries. - Run dry-run first: use the default (no --apply) to produce DREAMS.md and AUDIT.md, then manually inspect those files before allowing changes. - Review the shipped code: the package has no published homepage or known owner. I found multiple code-quality issues in the audit script (e.g., typos such as a stray space before strftime and an incomplete statement toward the end of the file) that will likely cause runtime errors; fix or review the code before using --apply. - No network exfiltration seen: the scripts operate locally (SQLite and markdown files) and contain no obvious HTTP/network calls or credential exfiltration. Still, because the source/owner is unknown, review the full scripts for any later modifications before trusting them. - Prefer local execution under controlled conditions: run the scripts as your user (not root), inspect outputs, and optionally run in a sandbox or VM if you want extra isolation. If you want, I can: - point out specific lines to fix in the audit script so it runs cleanly, or - produce a short checklist of the exact backup commands to run before first execution.

Type: OpenClaw Skill Name: memos-dreaming Version: 1.0.0 The memos-dreaming skill bundle is a legitimate utility for consolidating and auditing an agent's local memory. It processes data from a local SQLite database (~/.openclaw/memos-local/memos.db) and markdown logs in the workspace to generate a long-term MEMORY.md file. The scripts (memos_dreaming.py and memos_dreaming_audit.py) use standard libraries, contain no network activity or obfuscation, and follow the logic described in the documentation. No evidence of malicious intent or prompt injection was found.