obsidian-life-memory

obsidian-life-memory-skill

Transform your OpenClaw workspace into a structured, navigable memory system using plain Markdown, with official Obsidian desktop + official Obsidian CLI as the runtime path.

⚠️ Important: This skill now assumes the official Obsidian app is installed and running.

The shipped bin/obsidian-cli is not a standalone vault editor anymore — it is a thin compatibility adapter for OpenClaw that forwards commands to the official Obsidian 1.12.7+ CLI while preserving the older command shape (read path=..., daily:read, search-content, etc.).

This skill provides persistent personal memory for AI agents through wikilinked notes, daily logs, backlink traversal, and graph-based context retrieval.

---

What It Does

This skill helps agents maintain continuity across sessions by organizing information into an Obsidian vault with:

• Daily notes — Timestamped logs of events, decisions, and context
• Wikilinked entities — People, projects, and concepts as connected nodes ([[Project Name]], [[Person]])
• Long-term memory — Distilled summaries from daily noise
• Automatic context injection — Pre-prompt hook loads relevant context at session start
• Auto-categorisation — gizmo-curate files raw text into the right vault location automatically

---

gizmo-curate: Auto-Categorisation

gizmo-curate is a companion tool that uses an LLM to auto-categorise raw text and file it into the correct vault location. Pass raw text → LLM decides path, title, and tags → writes via obsidian-cli. No manual path decisions needed.

# File a person automatically
gizmo-curate "Alice Smith is a contractor at Acme Corp, [email protected]"
# → creates People/alice-smith.md ✅

# Append to an existing project note
gizmo-curate "Project Alpha: v2 integration in progress, review scheduled Friday"
# → appends to Projects/project-alpha.md ✅

# Preview without writing
gizmo-curate --dry-run "some text"

# Pipe mode
echo "some text" | gizmo-curate

Speed: ~1.2s per call.

Decision rules

Integration points

• Nightly cron — auto-curates new entities found in daily notes
• Manual — call gizmo-curate any time for instant filing
• Coding agents — available on PATH so agents can save project context

Binary: /usr/local/bin/gizmo-curate · Requires: LLM API key (configured in OpenClaw runtime env)

---

obsidian-cli — Compatibility Adapter over the Official CLI

The skill ships bin/obsidian-cli and installs it to /usr/local/bin/obsidian-cli, but that file is now a compatibility adapter.

Under the hood it calls the official Obsidian CLI from the desktop installer (tested with 1.12.7) and preserves the older OpenClaw-friendly command shape used throughout the vault and hooks.

The golden rule: all vault reads and writes go through obsidian-cli. Never use raw file tools on vault files.

Runtime requirements

• Official Obsidian desktop installed (1.12.7+ recommended)
• Official Obsidian CLI available (bundled with the installer)
• Obsidian app running
• In root/headless environments: start Obsidian with --no-sandbox and usually DISPLAY=:99

Installation

cp bin/obsidian-cli /usr/local/bin/obsidian-cli
chmod +x /usr/local/bin/obsidian-cli

Command Reference

Reading

obsidian-cli read path="MEMORY.md"
obsidian-cli read path="People/alice.md"
obsidian-cli daily:read                         # today's daily note

Writing & Editing

\n is interpreted — all write commands use printf '%b' internally, so \n becomes a real newline.

obsidian-cli daily:append content="## 14:30 UTC\n- Meeting notes here"
obsidian-cli append path="MEMORY.md" content="\n## New section\n- important fact"
obsidian-cli create path="People/alice.md" content="# Alice\nEngineer at Acme Corp"
obsidian-cli edit path="MEMORY.md" find="🟡 PENDING" replace="✅ DONE"

Browsing & Searching

obsidian-cli list folder="People"
obsidian-cli search query="project"             # filename search
obsidian-cli search-content query="invoice"     # full-text with context lines
obsidian-cli status                             # vault overview

Moving

# Renames file AND updates [[wikilinks]] across the vault
obsidian-cli move path="People/old-name.md" to="People/correct-name.md"

---

Vault Taxonomy

workspace/
├── IDENTITY.md              # Who the agent is (name, vibe, emoji)
├── USER.md                  # Who the human is (preferences, context)
├── SOUL.md                  # Behavioral guidelines and boundaries
├── MEMORY.md                # Long-term distilled knowledge
├── HEARTBEAT.md             # System status and maintenance config
├── AGENTS.md                # Workspace rules and tool usage
│
├── memory/                  # Daily notes (YYYY-MM-DD.md)
├── People/                  # Real people the human interacts with
├── Projects/                # Active projects with backlinks
├── Places/                  # Real locations
├── Finance/                 # Financial facts, mortgages, investments
├── Vendors/                 # Companies and services
├── Notes/                   # Loose notes and catch-all
└── Archives/                # Completed/cold projects

---

Session Startup Sequence

obsidian-cli read path="SOUL.md"
obsidian-cli read path="USER.md"
obsidian-cli daily:read
# Main sessions only:
obsidian-cli read path="MEMORY.md"

---

Wiki Pattern (Karpathy llm-wiki)

The vault implements the llm-wiki pattern — a persistent, compounding knowledge base maintained by the LLM. Instead of re-deriving answers from raw sources on every query, the agent incrementally builds and maintains a structured wiki of interlinked Markdown pages.

Three workflows

Ingest — You send a URL, PDF, or text. The agent reads it, creates a Sources/YYYY-MM-DD_slug.md summary page, updates relevant entity pages (Projects/, People/, Places/), updates index.md, and appends to log.md. A single source can touch 5–15 pages.

Query → file back — When the agent produces a substantive synthesis (analysis, comparison, research summary), it offers to file the result as a wiki page. Your explorations compound in the knowledge base, not just disappear into chat history.

Lint — Weekly (or on-demand) health check: dead wikilinks, orphan pages, important stubs, stale status items. Runs as a scheduled cron job.

Key files

Vault taxonomy (extended)

workspace/
├── index.md                 # Wiki catalog — all pages with one-liners
├── log.md                   # Append-only ingest/query/lint history
├── Sources/                 # External source summaries (one per source)
├── Projects/                # Active projects
├── People/                  # Real people
├── Places/                  # Real locations
...

---

Token-Optimized Pre-Prompt Hook

hooks/obsidian-preprompt.js handles two injection paths:

Session start (agent:bootstrap): injects governance files + daily snapshot once per session. It also automatically loads any files specified in the `OBSIDIAN_OPTIONAL_CONTE