cd /news/ai-agents/a-self-organizing-obsidian-vault-pow… · home topics ai-agents article
[ARTICLE · art-32398] src=github.com ↗ pub= topic=ai-agents verified=true sentiment=· neutral

A self-organizing Obsidian Vault powered by autonomous AI agents

A developer has designed a self-organizing Obsidian vault powered by autonomous AI agents that automate curation, synthesis, and task extraction while maintaining strict separation between work, personal, and resource folders. The system uses a continuous loop of capturing raw data, processing it into actionable insights, and pruning waste to keep the vault lean and high-signal. This approach aims to augment human understanding by offloading labor to AI agents while preserving the human's final understanding.

read4 min views1 publishedJun 18, 2026

This document defines the core philosophy, workflows, and operational logic of this Second Brain. It is a living document, updated as the system and its tools evolve.

Honest & Objective Thinking: This is a second brain — the work is finding the best solution, not an agreeable one. Agents challenge weak work (including another agent's and their own), object plainly when something is wrong, hold their position under pushback unless genuinely proven wrong, and ground claims in verified evidence. Never flatter; never pass average work to keep the peace.Lite over Large: Keep the vault lean and high-signal. Deletespent intermediates(a bookmark once curated, a draft once synthesized) — git preserves history, so deletion is safe, and agents shouldn't burn tokens on dead files. Keep only durable outputs; a big graph is a vanity metric.Strict Separation (The Hard Wall): Keep Work, Personal, and Resources folders strictly separated to prevent context bleed. However, cross-domain reasoning is enabled via thetype

frontmatter property (e.g.evergreen

,synthesis

), allowing cross-Area insights.Link-First Architecture: Knowledge value lives in the connections ([[wikilinks]]

), not just the content.Agent-Augmented, Not Agent-Led: AI agents (Claude/Antigravity) automate the labor (curation, synthesis, linting) while the human retains the final understanding.

: Raw captures, web clippings, and fleeting notes.00-inbox/

: Areas of responsibility and active efforts related to professional life.01-work/

: Areas of interest and life management related to personal life.02-personal/

: Reference library and topics of interest not tied to a specific responsibility.03-resources/

: Inactive areas or resources; cold storage.04-archive/

: Metadata, templates, attachments, and system documentation.99-system/

The core engine of the Second Brain is the continuous loop of capturing raw data, processing it into actionable insights, and pruning the waste.

flowchart TD
    %% Nodes
    Inbox["00-inbox (Raw Bookmarks)"]
    Curate["curate-bookmarks (Skill)"]
    Draft["draft/ (Curated Notes)"]
    Synthesize["synthesize-drafts (Skill)"]
    Plan["synthesis/ (Strategic Plan)"]
    Extract["plan-to-kanban (Skill)"]
    Kanban["todo-kanban.md (Actionable Tasks)"]
    Execute["Build / Execute"]
    Prune["Delete Intermediates (Bookmarks & Drafts)"]
    
    %% Flow
    Inbox --> Curate
    Curate -->|Extracts signal| Draft
    Draft --> Synthesize
    Synthesize -->|Scientific Thematic Synthesis| Plan
    Plan --> Extract
    Extract -->|Deduplicates and appends| Kanban
    Kanban --> Execute
    Execute --> Prune
    
    classDef file fill:#2d2d2d,stroke:#555,stroke-width:1px,color:#fff
    classDef skill fill:#1a3f5c,stroke:#4a90e2,stroke-width:2px,color:#fff
    classDef action fill:#1e4620,stroke:#4caf50,stroke-width:2px,color:#fff
    
    class Inbox,Draft,Plan,Kanban file
    class Curate,Synthesize,Extract skill
    class Execute,Prune action

Capture: Raw material lands in00-inbox/

.Curate (: The agent reads the inbox items, extracts the core value ("what we can steal"), and moves them into acurate-bookmarks

)draft/

folder within a specific Area. The source is logged inprocessed-sources.md

.Synthesize (: The agent takes multiple drafts, analyzes them against each other using a scientific thematic matrix, and generates a unified Strategic Plan (synthesize-drafts

)synthesis/

).Action (: The agent reads the Strategic Plan, extracts the actionable tasks, deduplicates them, and appends them to the Area'splan-to-kanban

)todo-kanban.md

.Clean: Once the knowledge is durable and actionable, the spent intermediates (the original bookmark and the draft) are deleted.

: Interactively creates a new Area by challenging the idea, defining goals/scope, and scaffolding the required hub notes, Kanban board, and folders.init-area

: Validates new ideas, challenges their utility, and scouts for external resources/tools to build them.scout-idea

: Processes inbox items into actionable Area drafts.curate-bookmarks

: Synthesizes multiple drafts in an Area into a strategic "Global Plan" using scientific thematic synthesis.synthesize-drafts

: Reads a synthesis document's action plan and extracts action items into the Area's Kanban board, deduplicating them.plan-to-kanban

: Read-only knowledge-graph integrity check — brokenvault-linter

[[wikilinks]]

, orphaned notes, and missingsource

/captured_from

traceability. Never edits.: Headlessly reviews pending maintenance tasks and peer-reviews tools created by other agents.audit-maintenance

Review Loop:vault/99-system/maintenance/agent-kanban.md

is a Kanban board with swimlanesTodo / In Progress / Done / Archived. Every tool creation must be logged as a new card under** Todo**.** Session Check**: ASessionStart

hook surfaces each agent's own pending reviews at session start by calling.claude/hooks/pending-reviews.sh <Reviewer>

.Quality Control: A reviewchallenges and hardensthe other agent's work — judging whether it produces the best-quality output, not just whether it follows format. A pass is earned; weak tools are failed with concrete, required improvements.

Source of Truth: Agents must readCLAUDE.md

and this document before making structural changes.No Direct Writes: Agents write todraft/

folders or specific system directories, never directly into the core of an Area without confirmation.Traceability: Every agent-created note must include asource

orcaptured_from

field.Security Guardrails: Agents stay inside the project directory and never read/write/exfiltrate credential or secret paths. Enforced perCLAUDE.md §10

.

To ensure the system remains robust and documented, every major change triggers this checklist:

Sync: Document the new capability or structural shift.README.md

Log: Add a card underagent-kanban.md

Todo, assigned to the other agent.** Audit ARA**: Confirm that no "projects" folders were created.

── more in #ai-agents 4 stories · sorted by recency
── more on @obsidian 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
Live at https://your-agent.zahid.host
Get free account → Pricing
from €0/mo · no card required
LIVE [news/a-self-organizing-ob…] indexed:0 read:4min 2026-06-18 ·