Skip to main content

Selected work · Personal knowledge tool

Second Brain.

A numbered Obsidian vault with CLI workflows, Claude agent instructions, and a research-to-knowledge pipeline. Built to turn notes, transcripts, and links into material I can pull from when writing a spec, briefing a project, or choosing what to study next.

System sketch Raw capture becomes connected knowledge and project material.

Obsidian gives me the readable surface: notes, backlinks, graph view, and task views. The CLI gives agents a way to search, read metadata, and update the vault without loading every Markdown file.

Context

How I went from a notes folder to a working knowledge system.

The repo started from a personal need: keep track of what I know, what I am learning, and what might become useful later. That includes business research, technical learning, games, project decisions, prompts, tables, and loose notes from the day. I care less about the archive existing and more about whether it can change a brief, a decision, a task, or the next thing I choose to study.

That is why the vault has clear layers. Inbox captures stay raw. Research notes preserve source material. Knowledge notes hold concise synthesis in my own words. Resources keep reusable artefacts. Projects apply all of that context to active work, whether it is freelance work, a personal project, or something I am studying in depth.

The same pattern maps cleanly to company work. A sales team could feed call transcripts into the system, extract the moves that actually help deals close, and turn those into onboarding material for the next seller. Support notes could become product evidence. Process notes could become training material instead of disappearing into a shared drive nobody opens.

Vault snapshot

The folder structure does some of the thinking.

The numbered layers are not decoration. They keep capture, project work, source material, synthesised knowledge, reusable assets, and system rules from collapsing into one folder called notes.

Current high-level architecture of the Second Brain vault.
second-brain
├── 00-inbox/              # Raw captures, unprocessed
├── 01-projects/           # Active work (.paused/ for paused)
│   └── {project-name}/
├── 02-personal/           # Personal notes by domain (career, finance, ...)
├── 03-research/           # Source summaries and references
│   └── {domain}/
├── 04-knowledge/          # Refined, atomic ideas
│   └── {domain}/
├── 05-resources/          # Reusable artefacts (shared/ is cross-domain)
│   ├── {domain}/
│   └── shared/
├── archive/               # Processed inbox and completed projects
├── design/                # Local Design Canvas app for visual artefacts
├── system/                # Architecture, workflows, tools, learnings
├── tasks/                 # Promoted task notes by status
│   ├── 00-todo/
│   ├── 01-doing/
│   ├── 02-waiting/
│   └── 03-done/
└── templates/             # Note templates and scaffold folders

Operating loops

Capture, research, extract.

Three steps that keep the vault from becoming a read-only archive. Each one feeds the next.

Capture raw inputs

Loose thoughts, links, transcripts, and half-formed questions can land in the inbox before I know what they are. I kept that layer intentionally rough.

Research source material

URLs, videos, calls, market questions, and documents become structured research notes with source metadata, status, summaries, key ideas, and review hooks.

Extract reusable knowledge

knowledge/extract keeps the pieces that changed how I think: principles, patterns, methods, concepts, questions, resources, or project inputs I can use again.

Operating material

What each folder is actually for.

One well-extracted note that reframes a client brief is more useful than a full archive nobody reads. The folder structure makes that distinction visible.

Workflow specimen Research is one layer, not the whole system.

The important decision is separation. Source summaries, personal synthesis, reusable assets, and project execution each have a different job, so they get different places to live.

00-inbox raw thought, link, note, transcript, question
01-projects applied work, decision, task, deliverable
03-research source-grounded summary with metadata
04-knowledge concise idea in my own words
05-resources prompt, table, reference, reusable artefact

Decisions

Three choices that made the vault usable.

  1. Keep capture raw, then process deliberately.

    Chose
    Keep 00-inbox loose enough for fast capture, then route material into research, knowledge, resources, or projects only after I have looked at it properly.
    Trade-off
    A messy inbox is tolerated on purpose. The trade-off is periodic triage, but the rest of the vault stays clear: raw thoughts do not compete with validated knowledge or active project work.

  2. Keep research and knowledge separate.

    Chose
    Let 03-research hold source summaries and references. Promote only personal synthesis into 04-knowledge as principle, pattern, method, concept, question, or map notes.
    Trade-off
    It adds one more step after summarisation. I chose that friction because a summary is still source material. It becomes knowledge only after I have judged it, rewritten it, and linked it back into the vault.

  3. Design for humans and agents at the same time.

    Chose
    Use Obsidian as the human interface for reading, linking, graph navigation, and task views, while agents use the CLI and committed Markdown instructions to work without loading the whole vault into context.
    Trade-off
    The setup is more constrained than plain files plus chat. I accepted that because Obsidian stays readable for me, while the CLI gives agents a way to work without dragging the whole vault into context.

Artefacts

What exists.

Personal tooling, so the proof is mostly in CLI command docs, Obsidian templates, and workflow rules rather than public screenshots.

Private walkthrough

Want to see how the system behaves in real work?

I can walk through the Obsidian surface, graph structure, CLI workflow, Research + Extract Knowledge loop, and how the same pattern would work inside a company without exposing private notes.

Book the walk-through