aDNA Startup First Hour

One hour, four stages, one project-aware AI agent. By the end you have a working governance layer — CLAUDE.md, STATE.md, and one ADR — that turns every future agent session from "let me re-explain the project" into "continue where we left off." Skim the stages below, then open the terminal and run the clock.

Who this is for

Startup CTOs and founding engineers on a 10–50 person team using AI agents heavily. You ship weekly, hire monthly, and your institutional knowledge is already living in Slack threads and three engineers' heads. The goal of this hour is to stop that leak — not to adopt the full triad today.

The four stages

Each stage names a time budget, an action, and the governance file it produces. Work them in order; don't skip ahead. The file produced in one stage is the input the next stage reads.

Minutes 0–10 · Clone and install

Fork the base template, clone it into your repo root, and open it in your editor. The base gives you the triad (what/, how/, who/), the 14-entity base ontology, and a working CLAUDE.md scaffold.

Minutes 10–25 · CLAUDE.md

Edit the root CLAUDE.md so it names your project, your voice, and the standing orders you care about. When you point an agent at your repo, this is the first thing it reads.

Minutes 25–40 · STATE.md

Write a one-paragraph operational snapshot: current phase, active blockers, next steps. This becomes the file every new hire and every new agent session reads first — your rolling sprint status in agent-addressable form.

Minutes 40–60 · First ADR

Capture one decision that currently lives in a Slack thread. Create what/decisions/adr_001_<short-slug>.md with context, decision, and consequences. Future agents — and future hires — will read this instead of re-asking.

Minutes 0–10 · Clone and install

Start from the base template (Agentic-DNA on GitHub) or the quick-fork flow in the base workspace guide. Copy the .adna/ template directory into your repo root so agents find it at a known path. No code changes yet — this stage is about getting the scaffold on disk and the directory tree in place.

  • Produced: base CLAUDE.md, MANIFEST.md, and the what/, how/, who/ triad directories.
  • Checkpoint: running ls what/ how/ who/ shows three populated directories; cat CLAUDE.md shows the generic standing-orders template.

Minutes 10–25 · Make CLAUDE.md yours

Open CLAUDE.md at the vault root and edit three things: the project name and one-line description in the opening header, the voice / tone paragraph (how you want agents to write), and the standing orders specific to your team (merge-freeze windows, review policies, default branch). The full Your First CLAUDE.md tutorial is the 20-minute long-form version of this stage if you want to go deeper.

  • Produced: a customized CLAUDE.md naming your project and your agent-work norms.
  • Checkpoint: open a fresh agent session pointed at your repo; the agent's opening message reflects your project name and voice rather than the generic template copy.

Minutes 25–40 · Seed STATE.md

Write a short operational snapshot — the shortest useful version is one paragraph: "We are in Q2 of $PROJECT. Current focus: $FEATURE. Blockers: none / <blocker>. Next up: <the 2-3 things that matter this week>." Keep it under 30 lines. The governance files concept page explains the four root files and why STATE.md is the "what's happening right now" slot.

  • Produced: STATE.md with current phase, blockers, and next steps.
  • Checkpoint: a new hire — or a new agent session — reading CLAUDE.md + STATE.md can answer "what is this project and what are we working on this week" without asking anyone.

Minutes 40–60 · File the first ADR

Pick one decision that currently lives only in Slack and write it as an Architecture Decision Record. Create what/decisions/adr_001_<slug>.md with three sections — Context, Decision, Consequences. Two paragraphs per section is plenty. Future agents and future hires read this instead of re-excavating the Slack thread. The knowledge-graph concept page explains how ADRs cross-link into the rest of the vault.

  • Produced: what/decisions/adr_001_*.md — the seed of your decision log.
  • Checkpoint: git commit, push, and open a fresh agent session. Ask "why did we choose <X>?" — the agent finds the ADR without being pointed at it.

End of hour · What changed

Three files exist that didn't exist at minute zero, and together they move the project from "ambient knowledge" to "agent-addressable knowledge." The next 60 minutes of your team's AI-assisted work reads these files before it generates anything.

Vault is legible to an agent

CLAUDE.md + STATE.md means a fresh agent session opens your project, reads two files, and knows where everything is. No per-session re-explanation.

Decisions stop evaporating

One ADR is the seed of a decision log. Within a month the three-founding-engineer bus factor starts breaking down — the "why" lives on disk.

Onboarding has an entry point

When your next hire starts, they read STATE.md and walk the ADR list. No Slack archaeology, no "ping me if you have questions."

Self-reference: this vault ran this hour

The aDNA.aDNA vault itself went through its own version of this hour on 2026-04-13: a forked base template, a customized CLAUDE.md, an initial STATE.md in how/, and a first ADR captured under what/decisions/. The session log for that bootstrap is visible in how/sessions/history/. Read through it to see the same hour play out on a vault you can browse.

Hour two and beyond

Don't scale prematurely. Run the vault for a week on just three files before adding more structure. When a second decision needs capturing, add adr_002. When the work gets bigger than a session, graduate to a mission file. The startup persona page sketches the 3–6 month trajectory.

Next Steps

Tutorial: Your First CLAUDE.md

The long-form tutorial for the 10–25 minute stage. Use it to go deeper when you have 20 minutes to invest in the agent-orientation file.

Governance Files

The four root-level files (CLAUDE.md, MANIFEST.md, STATE.md, AGENTS.md) and what each one does. Reference this when you graduate past hour one.

Startup Persona

Quick-reference adopter profile: pain points, typical ontology extensions (api_spec, onboarding_checklist, customer_feedback), and deployment pattern.

Tutorial: Design a Mission

Hour one is individual discipline; hour two is team coordination. When you have a sprint-sized goal, a mission file turns it into claimable objectives.