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. 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. Delete spent 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 the type 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 in 00-inbox/ . Curate : The agent reads the inbox items, extracts the core value "what we can steal" , and moves them into a curate-bookmarks draft/ folder within a specific Area. The source is logged in processed-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's plan-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 — broken vault-linter wikilinks , orphaned notes, and missing source / 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 swimlanes Todo / In Progress / Done / Archived . Every tool creation must be logged as a new card under Todo . Session Check : A SessionStart hook surfaces each agent's own pending reviews at session start by calling .claude/hooks/pending-reviews.sh