BETTER INSTRUCTIONS. MORE DELIBERATE EXECUTION.

Plan the job
before you
run the agent.

Turn a complex job into clear execution instructions. Define the outcome, set boundaries, plan for failure — then hand the runbook to your agent.

No account. No API calls. Your draft stays on your device.

RUNBOOK / PRODUCTIZE A PROTOTYPE01—05
01

Define the outcome

A polished public-facing product, with existing features preserved.

02

Bound the work

Improve copy and user flows. Preserve routes, architecture, and data.

03

Set approval gates

Implement independently. Return before publishing.

04

Plan for recovery

Try one reasonable alternative. Keep completed work.

05

Make done testable

Primary flows work. No placeholders. Changes validated.

Customize this recipe
FROM A REQUEST TO AN EXECUTION SPECIFICATION

Give the job a clear boundary.

ORDINARY PROMPT

“Improve my website.”

The outcome, permissions, and stopping point are left open.

AGENT RUNBOOK

Objective + Context + Scope + Constraints + Autonomy + Approval + Recovery + Done

Decide what good execution looks like before it begins.

A USEFUL STARTING POINT

Choose the job. Make it yours.

All 8 recipes ↗

Designed for capable AI agents, including GPT-6 Astra.

Customize here. Copy the runbook. Execute in your own agent environment.