Walkthrough

What it looks like, from a sentence to a working build.

Six steps, in order, with the button you click at each one. The whole review usually takes a few minutes, and you can stop after Step 5 — the rest happens in your own coding agent.

The end-to-end flow

Advance at your own pace, or let it play. Every step is written out below the illustration, so nothing depends on seeing the animation.

  1. Step 1

    Say what you are building

    Type it the way you would tell a colleague. No account, no repository, no source code — and no special vocabulary. If you would rather not write anything, Pick what you are building walks you through a short menu instead.

    Click Review this build

    Nothing leaves your browser at this point.

    agentdirective.org

    What are you asking your AI to build?

    A support tool that reads customer emails and automatically issues refunds.
    Review this build →Pick what you are building
  2. Step 2

    Check what it found

    Recognition is deterministic — the same words always produce the same result. Every item is a checkbox you control: untick anything that does not apply, tick anything that was missed. What you confirm here is what drives the rest.

    Click Continue

    Correct anything here — your corrections drive which safeguards apply.

    Confirm understanding

    We detected 4 characteristics

    • AI component reads customer content
    • Untrusted input from customers
    • Money movement — refunds
    • Customer data
    Continue →
  3. Step 3

    See which safeguards apply, and why

    Your confirmed characteristics select versioned Control Packs through a rule engine, not a model. Each one shows the characteristic that triggered it, so you can check the reasoning rather than trust it.

    Same description, same packs — every time.

    Matched Control Packs
    AUT-001 · Bounded AutonomyApplied because an automated action moves money.
    AI-001 · Untrusted Content and Tool BoundariesApplied because a model reads content it did not author.
    OBS-001 · Traceable Consequential ActionsApplied because a consequential action runs without a person present.
  4. Step 4

    Answer only what needs you

    These are the decisions a coding agent should never invent: limits, approvals, who may do what. Answer the ones you know. Leaving one open is a real answer — it is recorded as unresolved so your agent asks you instead of guessing.

    Click Review my answers

    An unanswered decision is never recorded as confirmed.

    Decisions · 1 of 6

    What is the maximum refund an automation may issue without human approval?

    • Any amount
    • Up to a set limit — $500
    • Not decided yet
    Review my answers →
  5. Step 5

    Take the result to your agent

    You get files your coding agent can read, and a copyable prompt that carries the whole directive as text — for pasting straight into a chat window. Copy agent prompt is the one most people want.

    Click Copy agent prompt

    Nothing is uploaded — the prompt is built in your browser.

    Your Agent Directive is ready
    ◆ agent-directive.json◆ AGENTDIRECTIVE.md◆ implementation-plan.md◆ control-status.json
    Copy agent promptDownload bundle
  6. Step 6

    Your agent builds it, and reports back

    Paste into Claude Code, Cursor, Codex — whatever you use. The directive opens by telling the agent to restate your boundaries and ask about anything still open, rather than choosing for you. As it builds, it records what it implemented and what it could not.

    Reported status is the agent’s claim — the optional review checks it against evidence.

    Your coding agent
    “Before I start: refunds are capped at $500 without approval, and every refund must be logged with the acting account. Two decisions are still open — who may approve above the cap, and how long logs are kept. How do you want those?”
    implemented · 7partial · 2unresolved · 2

Ready when you are

Try it with your own description.

It takes a few minutes and asks nothing of you but the description you already had in mind.