Memory Dream
v1.1.0Auto-consolidates agent memory files after sessions — like REM sleep for AI agents. Prevents MEMORY.md from growing unbounded by using an LLM to prune stale...
⭐ 1· 119·0 current·0 all-time
byMitch Davis@mitchelldavis44
MIT-0
Download zip
LicenseMIT-0 · Free to use, modify, and redistribute. No attribution required.
Security Scan
OpenClaw
Benign
high confidencePurpose & Capability
Name/description (auto-consolidate MEMORY.md using an LLM) align with what the code does: detecting session boundaries, reading memory files, optional lossless-claw summaries, recent transcripts, and calling a subagent LLM to rewrite and write back memory files. No unrelated credentials, binaries, or external downloads are requested.
Instruction Scope
SKILL.md and code are consistent but the runtime instructions and implementation access several places beyond the workspace file(s): it reads recent transcript (.jsonl) files from ~/.openclaw/<agentId>, optional lossless-claw summaries in the state directory, and writes a staging file (memory-staging.md) and modifies the configured memory files in the workspace. This behavior is coherent for consolidation but is broader than 'only MEMORY.md' (it draws on session transcripts and stores state under the user's home directory).
Install Mechanism
No install script or remote download is present; the package is plain TypeScript/JS source. There are no external URLs or extraction steps. The code will run as a plugin via the OpenClaw plugin system (normal for this ecosystem).
Credentials
The skill requests no environment variables, no credentials, and no config paths beyond per-agent state in the user's home directory and the agent workspace. Those accesses are justified by the feature (storing plugin state and reading transcripts/context).
Persistence & Privilege
The plugin is capable of autonomous invocation (default platform behavior) and will run background consolidations once trigger conditions are met; it stores state under ~/.openclaw/memory-dream/<agentId> and creates/clears a workspace staging file. It does not use always:true, nor does it modify other plugins' configs, but users should be aware it will autonomously modify workspace files when triggered.
Assessment
This plugin appears to do what it says: it uses your configured LLM to rewrite and tighten memory files, and it relies on session transcripts and optional lossless-claw summaries to decide what to keep. Before enabling: (1) back up MEMORY.md (or whatever files you configure) so you can review/revert automated edits; (2) review whether you are comfortable with a plugin reading recent .jsonl transcripts from ~/.openclaw/<agentId> and writing state under ~/.openclaw/memory-dream/<agentId>; (3) set conservative triggers (higher minSessions/minHours) if you want fewer automated runs; (4) choose an appropriate (and cost-appropriate) model in config; and (5) if you want to avoid any automatic writes, keep the plugin disabled and run consolidation manually. Overall the components are coherent and proportionate to the described purpose.Like a lobster shell, security has layers — review code before you run it.
latestvk97e749nydk6b8b64vk5w3xpnd847gx2
License
MIT-0
Free to use, modify, and redistribute. No attribution required.