What is aDNA?

aDNA (Agentic DNA) is an open standard for organizing what a project knows, so that people and AI agents can both find their way around it. Every aDNA project has the same shape. Learn that shape once and you can open any of them — including this site, which is one.

A note on the name. In genomics, aDNA usually means ancient DNA. This is not that. Here it stands for Agentic DNA, and the borrowing is deliberate: a genome is structure a cell inherits and reads, and these files are structure a project keeps and its agents read.

The problem

AI agents have the same trouble people do: finding the right file. With no shape to follow, an agent reads the wrong thing, or misses the thing that counts. So you explain the project again at the start of each session. Agents undo decisions they made last week. Work slips out of reach as the context window fills.

The problem is the filing, not the agent. The agent is able; it has nowhere to look. Most teams patch the gap with long READMEs and custom prompts, and none of that carries to the next session, the next agent, or the next teammate. aDNA is one open answer. Any team can adopt it, any tool can support it, and any agent can read it with no setup.

How aDNA works

aDNA gives a project three things:

• The Triad — three directories, in every project: who/ for people and governance, what/ for knowledge and decisions, how/ for operations and work. See one aDNA project and you know your way around the next.
• Governance files — five files that orient an agent at each level: CLAUDE.md, AGENTS.md, MANIFEST.md, STATE.md and README.md. The root CLAUDE.md is the front door: what the project is, how it is laid out, what the rules are, where to begin.
• Typed entities — 16 base types, among them missions, sessions, skills and templates. Each type uses the same frontmatter fields and the same naming, everywhere. Read one mission file and you can read any mission file, in any aDNA project.

What a project looks like

A small aDNA project — three directories, five governance files:

your-project.aDNA/
├── CLAUDE.md          # Agent entry point: purpose, rules, where to start
├── STATE.md           # Live snapshot: blockers, active work, next steps
├── MANIFEST.md        # Project overview and architecture
├── who/               # WHO — people, governance, coordination
│   └── governance/    # Roles, policies, vision
├── what/              # WHAT — knowledge, decisions, context
│   └── context/       # Curated knowledge files agents load
└── how/               # HOW — operations, plans, execution
    ├── missions/      # Work decomposed into claimable objectives
    └── sessions/      # Per-session tracking and handoff notes

An agent that has seen one aDNA project knows this at a glance, before it reads a word of the content.

The 16 entity types

Every file states its type. Because the types never change, an agent can work in a project it has never seen, with nothing to set up first. The 16 base types are split across the Triad — 4 WHO, 5 WHAT, 7 HOW. Here is the whole set:

What a CLAUDE.md looks like

A CLAUDE.md is not a README. It is the agent's operating protocol, and the agent reads it first, every session. Instead of "work it out from the README", the project and the agent share one set of terms.

# CLAUDE.md — aDNA.aDNA

You are Rosetta — named after the Rosetta Stone, the artifact that decoded
Egyptian hieroglyphics by presenting the same text in three scripts. This
vault does the same: it presents the aDNA standard in three registers —
technical specification, operational practice, and plain-language explanation.

## Project Map

aDNA.aDNA/
├── CLAUDE.md     ← You are here — agent master context (this file)
├── STATE.md      ← Operational snapshot: current phase, blockers, next steps
├── what/         ← Knowledge objects, context library, lattice definitions
├── how/          ← Operations, sessions, missions, campaigns, skills
└── who/          ← Governance, community, coordination

## Standing Orders

1. Phase gates are human gates — never auto-advance between phases.
2. Every mission gets an AAR before marking it completed.
3. Upstream spec is source of truth — cite adna_standard.md for normative claims.

Before and after

The contrast below is the general pattern the standard is built against — not a case study, and no measured project is being described.

Without aDNA: what the project knows is spread across Notion, Drive and Git. Each session opens with a pasted summary that is already out of date. Last month's decisions get argued again. A new teammate has to work out where everything sits before they can start.

With aDNA: the same project has a what/context/ library, a STATE.md that names the current priorities and blockers, and a how/missions/ directory where work is split into pieces someone can claim. An agent opens CLAUDE.md, reads the context it points to, and starts in the right direction — in that same session.

See for yourself

aDNA is not a concept deck. It is a standard you can clone and read today. The public image at github.com/aDNA-Network/aDNA is a real aDNA workspace. One command gives you the standard, the skills and the templates. A fresh clone also offers to set up a complete Home for an agent. Open the files yourself:

• CLAUDE.md — the workspace operating protocol an agent loads first
• .adna/ — the full aDNA standard, embedded and ready to clone
• .adna/how/skills/skill_onboarding.md — the first-run recipe that orients a new project
• .adna/how/skills/skill_project_fork.md — how a new .aDNA project is created
• .adna/how/templates/template_workspace_claude.md — the router template every workspace starts from
• .adna/how/templates/template_home_claude.md — the governance template a fresh clone uses to bootstrap a polished Home (shipped at v8.0)
• .adna/how/templates/template_node_adna_exemplar/ — the themed exemplar Home the bootstrap offers out of the box (shipped at v8.0)

This site is itself an aDNA vault. The shape you are reading about is the shape that produced it. Clone the image and open it in Obsidian, in VS Code, or on GitHub — every directory and every frontmatter field is one of these ideas at work.

The three-question test

A well-built aDNA project lets any agent answer three questions at once, without asking:

1. What is this project? — CLAUDE.md and MANIFEST.md at the root.
2. Where does it stand? — STATE.md: blockers, active work, next steps.
3. Where do I start? — the open mission in how/missions/, or the nearest AGENTS.md.

If your project answers all three within ten seconds of reading, it is ready for aDNA.

Explore further

• The Triad — the structure underneath it all: who, what, how
• Governance Files — CLAUDE.md, AGENTS.md, STATE.md, and what each is for
• Get Started — set up your first aDNA project in three steps
• Tutorial: Create Your First CLAUDE.md — hands-on, 20 minutes
• The Convergence Model — how aDNA narrows what an agent loads at each level of the work