Skip to main content

AI Productivity Kit

Rules, prompts, and config files you copy into your repo so Cursor (and Copilot / Antigravity) plan before coding, ask one question at a time, and pause for security when it matters. Use it for any project—web apps, APIs, mobile, data tools, CLIs—not just websites.

:::info Any stack, any project The kit works the same whether you're building a website, a Python app, an API, a CLI, or an internal tool. The workflow (plan → spec → code → security) is the same; only your tech stack and optional design system change. :::

Quick start​

:::tip (5 minutes) Step 0 — Get your stuff together
Have your project folder ready (new or existing). If you have a UI, have your design system name and docs handy. See Prerequisites.

Step 1 — Get the files
Download ai-kit-only.zip from Releases and unzip into your project root. Or copy everything from kit/ into your project root.

Step 2 — Fill in your config
Open docs/ai/ai-config.md and fill in your platform type and project info. If you have a UI or design system, add its name, docs URL, and import pattern (e.g. import { Button } from '@/components/ui'). Back-end-only or CLI projects can leave design system as TODO.

Step 3 — Open Cursor (or Antigravity) and paste the Session Kickoff
Start a new Composer session and paste the Session Kickoff prompt. In Google Antigravity, open the repo there; rules load from .agent/rules/ automatically.
Full text: Session Kickoff reference.

Step 4 — Type your first task
The AI will clarify one thing if it needs to, then plan before coding. :::

What it does​

  • Plan first — Clarify the problem and acceptance criteria before code.
  • Two modes — Plan mode (planning, PM, design, security) and Agent mode (code, tests, safe diffs). In Cursor, toggle with Shift+Tab. In other editors, type Switch: SPEC or Switch: IMPLEMENT.
  • One question at a time — The AI asks exactly one question when it needs input, then continues.
  • Advisories — Every response starts with a short block: route (Plan/Agent), recommended model, context risk, and when to switch or start a new thread.
  • Security stop gate — For high-risk work (auth, permissions, exports, uploads, integrations, confidential data), the AI asks one security question and stops until you answer.

How Plan mode and Agent mode work together​

Get started​

  1. What is this? — Overview and big idea.
  2. Install (by copying the kit) — Copy the kit/ into your repo and fill placeholders.
  3. Daily workflow (SPEC-first) — Session Kickoff, Context Pack, Router.
  4. Cursor modes — Chat vs multi-file vs small edits.
  5. Context windows — Overload, rot, and reset.
  6. Model switching — When to use fast vs reasoning vs best-coding.

Preview the kit files: See the folder structure and exact text of each file in Reference → Kit file preview. To understand when and how to use agents (e.g. @Security, @PM), see Context pack → Agents.

Use the Docs menu on the left to browse the full guide.