Learning,
by building.

The loop, step by step

Each step has a skill

A skill is a short instruction file that your AI tool loads when you need it. Each one asks for your own idea, example or guess first, then helps you take it further. Say “just do it” when you only want the task done.

1. 01 / Imagine

   What could work better?

   • /frame-the-problem Turns a vague complaint into one specific problem and a first thing to try.
   • /brainstorm You list a few ideas first, then it adds many more, from a saved prompt to a small app.

2. 02 / Make

   Build something small & useful.

   • /build-brief Picks the simplest version that could work and writes down what it must do and how you will check it.

3. 03 / Check

   Does it actually do the job?

   • /prove-it Checks the result against the source, the real number or a rerun, never against the AI’s opinion of itself.
   • /expert-council Gives a plan or draft to three to five critics who each say what worries them.

   /stuck-or-pivot When it isn’t working: run one test, then fix it, simplify it or change approach.

4. 04 / Repeat

   Make it part of your work.

   • /make-it-repeatable Saves what worked as a prompt, template or checklist you can reuse and share.

A commons that connects

Every idea links to the work that uses it

Each concept card links to the products that use the idea, the resources that teach it, the skills and blueprints that practise it, and the names it goes by in other fields. A project manager who runs risk reviews already knows half of what a pre-mortem asks.

• 44 ideas
• 8 skills
• 10 blueprints
• 97 resources
• 1025 connections

Field notes / one idea and its links

Context engineering

Choosing and arranging everything the AI can see, such as instructions, examples, files, past messages and connected data, so it has what it needs for the task and little that distracts it.

You may know it as briefing a contractor (project management) or mise en place (cooking).

• Concept: Prompting (garden)
• Concept: Context window (garden)
• Concept: Prompt engineering (garden)
• Concept: Memory (garden)
• Competency: Describe the task
• Concept: Projects and knowledge (garden)
• Concept: Skills (garden)
• Blueprint: Hire and onboard an AI copilot for one piece of work (garden)
• Blueprint: Let AI ask before it plans (garden)
• Blueprint: Teach your AI assistant your team's jargon, one confirmed term at a time (garden)
• Skill: /build-brief (garden)
• Resource: Introduction to Claude Cowork (garden)
• Resource: Claude Help Center: Introduction to prompt design (garden)
• Resource: Gemini API prompt design strategies (garden)
• Resource: How to Write a PRD: A Step-by-Step Guide for Product Managers (garden)
• Resource: Claude Cowork for PMs: My Self-Improving Productivity System (How I AI, with Daniel Blum) (garden)
• Resource: Compound Engineering (garden)
• Resource: Build your personal AI copilot (garden)
• Resource: Lenny's Data: free public starter pack (garden)
• Resource: How to build your PM second brain with ChatGPT (garden)
• Resource: Getting started with AI: Good enough prompting (garden)
• Resource: 12 Rules to Vibe Code Without Frustration (garden)
• Resource: Behind the Craft — Transcripts (garden)

Start small

Your first ten minutes

1. Pick one task you repeated this week and would like to stop doing by hand: a status update, a messy inbox, a plan you keep putting off.
2. Open the AI chat you already use. A free plan is fine.
3. Run one blueprint from start to finish. Turn messy notes into a status draft you can check takes about 25 minutes, and its first step takes five.
4. Check one line of the result against your own notes. Every blueprint here asks you to do that before you trust the output.

To load the skills into your own AI tool, follow Start here.