Field guide / sub-domain
6
sources in this field
Updated September 19, 2026
Current thesis
The shortest path to orientation.
This synthesis records claims and practices from the cited sources; reported outcomes and product capabilities have not been independently verified.
Knowledge tooling reduces friction between raw content and structured, agent-readable information through markdown-as-interface patterns. Ars Umbris treats knowledge bases like codebases: typed files with explicit relationships an agent can parse, rather than freeform notes. Repos bundle knowledge, type definitions, agent skills, MCP tools, UI projections, and instructions as local git-trackable files with dependencies declared via manifest (repo.yaml), making structure compound across sessions. Cross-repo references use explicit syntax (Note::Another Repo, source::au-base-types*), creating typed graphs analogous to imports. URL-based tools (Mintlify, Google's Open Knowledge Format) enable agents to query and edit markdown programmatically. Memory persistence (Gbrain, QMD+SQL) accumulates research across cycles. Self-hosted organizers (Siftly) and local graphs (FieldTheory) solve graveyard problems without API limits. Obsidian surfaces agent skills via file architecture; Cabinet, Tolaria, and ByteRover unify scattered markdown into indices. Design-system libraries package 2,000+ products as DESIGN.md files readable by Claude Code. Excalidraw (110K stars) enables agents to auto-generate architecture diagrams. Design philosophy leverages multiple projections—partial UI views—to reveal understanding of knowledge graphs too large for single comprehension.
Evidence board
Claims worth carrying forward01Ars Umbris treats knowledge bases like codebases: typed files with explicit relationships and conventions an agent can parse, rather than freeform notes, so structure compounds across sessions instead of decaying.
02A repo bundles six parts as local files: knowledge (notes/sources), types (definitions), skills (agent instructions), MCP tools (operations), projections (UI components), and agent instructions/profiles—all composable and git-trackable.
03Repos declare dependencies via a manifest (repo.yaml with a deps list), similar to package imports in code, letting one repo (e.g. au-tree-research) supply vocabulary, skills, and MCP tools that other repos reuse without duplication.
04Cross-repo references use explicit syntax: [[note::another-repo]] for wikilinks and source::au-base-types* for shared types, making dependencies explicit like imports—so a workspace of multiple repos reads as one typed graph.
05Core thesis: knowledge work needs its own type engine (analogous to compilers/linters in software) because agents can silently accumulate broken links, missing fields, and inconsistent structures across sessions—reminding the model of conventions isn't sufficient at scale; independent checks are required.
06Type definitions (YAML) specify required fields and valid reference targets (e.g., a claim type requiring a source::au-base-types field); when an agent writes a note missing that field, the engine flags a diagnostic the agent can read via tools and fix—though judging whether a source actually supports a claim still requires human/agent judgment.
07Ars Umbris positions itself as a 'meta harness': it connects multiple agent harnesses (Claude Code, Codex) via MCP adapters to one shared durable workspace state, so separate agent sessions with independent context can still read/write the same files and continue each other's work.
08Design philosophy (named after Giordano Bruno's 'shadows of ideas'): since a knowledge graph grows beyond what a person can hold in mind, the system relies on multiple projections (UI views) that each reveal a partial slice, and moving between them builds understanding of the whole.
Adjacent fields
Key voices
Latest evidence
Heinrich
Heinrich launches the early alpha of Ars Umbris, a malleable IDE that treats a knowledge base like a codebase—typed files, composable repos, cross-repo dependencies, and a type engine that gives agents compiler-like feedback on knowledge structure.
Matt Pocock
Humanlayer's 'show-me' skill (by @dexhorthy) is a markdown-defined toolbox of visual representations—pseudocode, call trees, component trees, file trees, Mermaid diagrams, diffs, and standalone HTML artifacts—used to make code and PR changes easier to understand at a glance.
Vox
A free library converts the design language of 2,000+ top products (Linear, Notion, ElevenLabs, Lamborghini) into DESIGN.md files—structured markdown that coding agents like Codex and Claude Code can read directly to build or align UIs with a chosen product's visual system.
Zack Kanter
Zack Kanter responds to Martin Casado's observation that AI-written PR/commit messages sound polished but are often confusing, proposing that agents be given explicit style constraints — git commit conventions, ASD-STE100 Simplified Technical English, and Google's developer documentation style guide — to fix the problem.
Marie Haynes
> This is really big news. Google introduced the Open Knowledge Format (OKF) - a standardized way to store information in a directory of markdown files. Makes it really easy to make a digital brain that agents can use. > > These files can serve as a living wiki. You can give agents the ability to q
Jonata Santos
> how to build anything rn: > > - get a hetzner, do, or hostinger vps