← Back to Skills Marketplace
oryanmoshe

Memory Dreaming

by Oryan Moshe · GitHub ↗ · v0.2.0 · MIT-0
cross-platform ⚠ suspicious
191
Downloads
0
Stars
0
Active Installs
2
Versions
Install in OpenClaw
/install memory-dreaming
Description
Autonomous memory consolidation for OpenClaw agents — like REM sleep. Periodically gathers signal from daily logs, session transcripts, and learnings; consol...
Usage Guidance
What to consider before installing: - This skill reads agent session transcripts (~/.openclaw/agents/...) and workspace files and will search for high-signal text including 'URLs' and 'API keys'. If your transcripts or notes contain secrets (API keys, passwords, tokens), the dream cycle may capture and persist them into MEMORY.md or into an Obsidian vault. If that vault is cloud-synced (Obsidian Sync, Dropbox, etc.), those secrets can be uploaded off your machine. - Recommendations before enabling automatic dreaming: - Inspect the included files (SKILL.md, scripts/*.sh, references/*) yourself — you already have them locally. Confirm you are comfortable with the exact paths it reads. - Keep obsidian.enabled=false unless you trust where the vault is stored. Prefer manual sync until you’ve validated outputs. - Run a single manual dream cycle in a controlled workspace (or test workspace) to see what the skill writes to MEMORY.md and dreaming-log.md before enabling cron. - Audit your session transcripts and workspace for secrets; consider clearing or redacting sensitive lines or reducing session retention if you don’t want those included. - Limit file-system permissions for the OpenClaw workspace or run the skill under a user that has access only to the intended directories. - Note: the scripts prompt a cron creation via OpenClaw — creating that cron enables periodic scanning; only create it if you want automatic runs. - Other notes: - The skill will update .learnings entries (mark as promoted) and write to MEMORY.md and dreaming-log.md — back up those files first if you need a restore point. - The gate script and prompt rely on file timestamps; behavior may vary on different OS/date implementations (the script attempts BSD/Linux fallbacks). If you want to proceed: disable obsidian sync or point vaultPath to a local, non-synced folder first; perform manual runs and review outputs; then consider enabling a cron with a long schedule and narrow source scope.
Capability Analysis
Type: OpenClaw Skill Name: memory-dreaming Version: 0.2.0 The skill implements an autonomous memory consolidation system that processes session logs and syncs knowledge to an Obsidian vault. It is classified as suspicious due to potential shell injection vulnerabilities within the agent instructions in `references/dream-prompt.md`, where the agent is directed to execute shell commands (e.g., `ls`, `cat`, `dirname`) using paths derived from user-controlled configuration and workspace directory names without sanitization. While the functionality is plausibly intended for memory management, the skill requires broad read access to sensitive session transcripts (`~/.openclaw/agents/main/sessions/*.jsonl`) and instructs the background agent to operate silently ('NO_REPLY'), which could mask unintended behavior if the shell commands are exploited.
Capability Assessment
Purpose & Capability
The declared purpose (consolidate daily logs, session transcripts, learnings; optionally sync to Obsidian) lines up with the code and instructions: scripts read memory files, grep session transcripts, scan workspace for task_plan.md, update MEMORY.md and (optionally) create/modify notes in an Obsidian vault. This access (workspace + session transcripts) is expected for a memory-consolidation skill, but it's broad (reads agent session history and arbitrary workspace files).
Instruction Scope
SKILL.md and the autonomous dream prompt explicitly instruct the agent to read HOME session files (~/.openclaw/agents/main/sessions/*.jsonl), full daily logs, .learnings files, and to extract 'URLs, API keys mentioned' and other entities. The skill runs as a cron job by default every 8 hours (configurable) and will write consolidated output to MEMORY.md and optionally to an external Obsidian vault. Those instructions can collect and persist sensitive data (secrets, API keys) and will run periodically without per-run human review unless the user disables the cron or obsidian sync.
Install Mechanism
No install spec or external downloads — it's instruction + scripts included in the repo. No remote code fetch; scripts are local and small. This is lower install risk.
Credentials
The skill declares no required env vars, but expects workspace/home paths (OPENCLAW_WORKSPACE defaulting to $HOME/.openclaw/workspace and explicit references to ~/.openclaw/agents/). It also recommends an absolute Obsidian vault path for syncing. The functional needs justify reading memory and plan files, but the explicit instruction to capture 'API keys' and similar entities is disproportionate and elevates risk because the skill will persist potentially sensitive tokens into MEMORY.md or an Obsidian vault (which may be synced to cloud services).
Persistence & Privilege
always is false and the skill is user-invocable; it proposes creating a cron in OpenClaw via setup-cron.sh which would enable periodic autonomous runs. The skill does not request to alter other skills or system-wide config. Cron-based periodic execution plus broad read access increases blast radius, but the privilege level requested is coherent with its purpose.
How to Use
  1. Make sure OpenClaw is installed (local or Docker)
  2. Run the install command in chat: /install memory-dreaming
  3. After installation, invoke the skill by name or use /memory-dreaming
  4. Provide required inputs per the skill's parameter spec and get structured output
Version History
v0.2.0
Plan tracking rewrite: concrete bash commands, file templates, step-by-step Obsidian sync logic. Dream prompt is now fully executable by an isolated agent with zero context.
v0.1.0
Initial release: 4-phase dream cycle, Obsidian vault sync, plan tracking, time-based gate, cron setup
Metadata
Slug memory-dreaming
Version 0.2.0
License MIT-0
All-time Installs 0
Active Installs 0
Total Versions 2
Frequently Asked Questions

What is Memory Dreaming?

Autonomous memory consolidation for OpenClaw agents — like REM sleep. Periodically gathers signal from daily logs, session transcripts, and learnings; consol... It is an AI Agent Skill for Claude Code / OpenClaw, with 191 downloads so far.

How do I install Memory Dreaming?

Run "/install memory-dreaming" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.

Is Memory Dreaming free?

Yes, Memory Dreaming is completely free, licensed under MIT-0. You can download, install and use it at no cost.

Which platforms does Memory Dreaming support?

Memory Dreaming is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).

Who created Memory Dreaming?

It is built and maintained by Oryan Moshe (@oryanmoshe); the current version is v0.2.0.

💬 Comments