What does this skill do, and when should you use it?
This skill teaches users to set up a persistent AI memory system built from three components: OpenClaw, Obsidian, and QMD. OpenClaw injects workspace files into the system prompt at session start, so the AI 'wakes up' already knowing its context; Obsidian visualizes connections between notes in Graph View; QMD provides on-device semantic search so relevant context is found without loading every file. The skill covers the file structure (MEMORY.md, daily logs, second-brain, directives), vault setup, plugin configuration, and a HEARTBEAT.md-driven routine for periodic memory maintenance. It comes from leondgarse's personal knowledge base repository Atom_notebook, which is itself a working instance of the method.
Guides the AI through: creating an Obsidian vault pointed at the workspace folder and enabling Graph View; installing and configuring the Dataview and Templater plugins with a daily note template and memory queries; running qmd collection add and qmd embed to build a semantic index, then mcporter call qmd.vsearch / qmd.query for retrieval; creating and maintaining the core files (MEMORY.md, memory/YYYY-MM-DD.md, second-brain/, directives/, HEARTBEAT.md, AGENTS.md, USER.md, SOUL.md); and following HEARTBEAT.md on a schedule to review recent logs, distill insights into MEMORY.md, and remove outdated entries.
- An individual who works with an AI assistant long-term and wants new sessions to pick up prior decisions and project context instead of re-explaining each time.
- An existing Obsidian user who wants to upgrade their vault into an AI-readable, AI-writable memory workspace.
- A user already running OpenClaw or similar session-start injection tooling who needs a proven file layout and maintenance workflow.
- Someone with a large note collection who doesn't want to load everything into the context window and needs QMD semantic search to recall relevant files.
- A user who wants the AI to act proactively between sessions (tidy logs, update long-term memory) via the HEARTBEAT.md scheduled-task mechanism.
How do you install this skill?
- The recommended structure stores API keys/credentials in plaintext in TOOLS.md; users should keep real credentials out of files injected into the system prompt and gitignored.
- MEMORY.md and memory files are injected into the system prompt; avoid writing sensitive personal data there.
- License is unknown; confirm terms before commercial use or redistribution.
- Core dependencies (OpenClaw/QMD) have missing/TBD documentation links and inconsistent repository attribution; mainland-China network reachability is unverified.
- Static review only — no commands were executed; verify QMD and mcporter command availability yourself.
- Shell / CLI
- Local filesystem
ObsidianOpenClawQMD CLIDataview pluginTemplater pluginmcporter (for QMD calls)
- Get skills/obsidian-openclaw-memory/SKILL.md from the leondgarse/Atom_notebook repository and place it in your Agent Skills directory (e.g. ~/.claude/skills/obsidian-openclaw-memory/). 2. Install Obsidian, OpenClaw, and QMD locally (installation instructions for QMD and mcporter are not provided in the source). 3. Open your workspace folder in Obsidian: File → Open Vault → Open Folder as Vault. The skill also references references/file-structure.md and assets/architecture.png, which are not included in the SKILL.md content shown; verify they exist in the repository.
tmp="$(mktemp -d)"
git clone --depth 1 https://github.com/leondgarse/Atom_notebook.git "$tmp"
mkdir -p ~/.claude/skills
cp -R "$tmp/skills/obsidian-openclaw-memory" ~/.claude/skills/
rm -rf "$tmp"Generated from the source repository and skill path; it copies only this skill's folder. If the author's install steps above differ, follow those first. To scope it to one project, replace ~/.claude/skills with that project's .claude/skills.
How do you use this skill?
Once installed, send your agent any of these to trigger it:
- Set up my workspace as an OpenClaw memory system
- Configure an Obsidian memory vault
- write this to memory
- Example trigger: 'Set up my workspace as an OpenClaw memory system' or 'Configure an Obsidian memory vault'. 2. Follow the skill to create the core files (MEMORY.md, memory/, HEARTBEAT.md, etc.). 3. Build the semantic index:
qmd collection add ~/clawd --name clawdthenqmd embed; re-runqmd embedafter adding significant new content. 4. Search from within a session:mcporter call qmd.vsearch query="...". 5. Explicitly ask the AI to 'write this to memory' for anything you want remembered; every few days follow HEARTBEAT.md to distill logs into MEMORY.md.
What are this skill's strengths and limitations?
- Provides a complete, concrete workspace file structure and templates (daily note template, Dataview query, HEARTBEAT example) you can copy directly.
- The core principle is clear and actionable: write important things to files rather than only saying them in chat, achieving cross-session memory via file injection.
- Distilled from the author's real personal knowledge base, with the README showing the full directory layout as a reference implementation.
- Clean three-component division of labor (injection, visualization, semantic search) avoids stuffing all files into the context window.
- Hard dependency on three external tools (OpenClaw, Obsidian, QMD); limited value if you don't use this stack.
- QMD's installation method and documentation are marked TBD in the source, so ecosystem maturity is uncertain.
- The repository has no declared license, creating licensing uncertainty before adoption.
- No topics, no tests, and no independent validation of the skill's effectiveness; HEARTBEAT scheduled maintenance depends on OpenClaw polling configuration the user must set up themselves.
- The referenced references/file-structure.md and assets/architecture.png are not included in the provided source, so completeness cannot be verified.
How does this skill compare with similar options?
Side by side with related skills; every score comes from the same FSRS standard.
| Skill | FS score | Stars | Last updated | License |
|---|---|---|---|---|
| Obsidian + OpenClaw Persistent AI Memory System this page | 46 · Use with care | ★ 95 | 11d ago | — |
| LLM Wiki — a second brain for AI agents | 58 · Recommended | ★ 121 | 1mo ago | MIT |
| Conversation Archivist | 49 · Use with care | ★ 16k | 1mo ago | MIT |
| Librarian: Knowledge Vault Steward | 50 · Use with care | ★ 32 | 4mo ago | MIT |
| Autonomous Research Filing | 52 · Use with care | ★ 16k | 1mo ago | MIT |
How did FollowSkills review this skill?
Documentation-only setup guide with no scripts or external writes, so inherent risk is low; however, the file structure recommends storing API keys/credentials in plaintext in TOOLS.md, and privacy handling is limited to a brief note on MEMORY.md loading. No confirmation, scoping, or rollback guidance — deducted.
Instructions are self-consistent and the Obsidian/QMD happy paths look plausible; but there are no tests, no error handling or failure feedback, and dependency availability (OpenClaw, QMD, mcporter) is unverified — deducted.
Target scenario (cross-session AI memory with Obsidian) is clearly described with decent file-lifecycle detail; but non-fit boundaries and trigger conditions are thin, and reachability of OpenClaw/QMD from mainland-China networks is unaddressed, with QMD docs listed as TBD — deducted.
Layered architecture (SKILL.md + references + suggested assets) with reasonable progressive disclosure; but license unknown, no versioning/changelog, unclear ownership, inconsistent paths (~/clawd vs ~/Atom_notebook), and README's GitHub link points to a different account (Samin12) — deducted.
Can plausibly help users assemble a memory workspace with structured file conventions; but it is a static guide, key OpenClaw/QMD commands lack reproduction evidence, and direct usability of the resulting setup is unverified — deducted.
Author-written docs only; referenced external resources are missing or inconsistent (QMD TBD, GitHub link to a third-party account); no tests, no third-party execution evidence, no independently reproducible material — deducted.
Open a dimension to read why it scored that way
Evidence confidence：Low — Mostly static review, author material or a limited demo; useful for discovery, not high-risk decisions.
See the full review method →