Smart Memory Keeper

What to consider before installing: - The skill is largely coherent: it is instruction-only and uses only the openclaw CLI and local files to implement a 3-tier memory system. There are no hidden downloads, no requested credentials, and no network endpoints. - Primary concerns to weigh: - It will read your most recent OpenClaw session(s) to build memory. That can expose any sensitive data contained in prior conversations. The SKILL.md says it will show you findings and ask for confirmation before writing tasks.md, but the read occurs up-front to generate that summary. - The installation requires you to append lines to global startup files (~/.openclaw/workspace/AGENTS.md and HEARTBEAT.md). Those files control agent startup behavior for the whole workspace; adding entries changes agent behavior beyond this single skill. The skill provides the exact content to append, but you should manually inspect and understand each line before applying. - Practical safety steps: 1. Back up ~/.openclaw/workspace/AGENTS.md and HEARTBEAT.md before making any changes so you can revert. 2. Manually copy the exact blocks from SKILL.md rather than running any automated 'install' that modifies global files without review. 3. Consider running the initial session scan only after sanitizing or removing particularly sensitive sessions, or run the one-time bootstrap manually and review the extracted summary carefully before approving writes. 4. Verify file permissions on ~/.openclaw/workspace/memory to ensure only your user can read/write those files. 5. If you prefer less blast radius, avoid appending to global AGENTS.md/HEARTBEAT.md and instead keep the memory rules in a local skill file and invoke the memory reads explicitly when needed. - Final note: because the skill modifies agent startup behavior and reads session history (sensitive data), treat it with caution and perform manual review/backups before applying its post-install steps.