Impact Lab — Intermediate Track · Practice on Your Own

Section 1

Your Environment, Rebuilt in 15 Minutes

The Impact Lab environment was Coder + Claude Code + GitLab Pages. Every piece has a free equivalent you can set up today.

Impact Lab Tool	Free Equivalent	How to Get It
Coder workspace	VS Code (local) or GitHub Codespaces	code.visualstudio.com · github.com/codespaces
Claude Code	Claude Code	Mac brew install claude Windows winget install Anthropic.Claude Either npm install -g @anthropic-ai/claude-code Requires a Claude.ai account · claude.ai/download
GitLab Pages	Vercel or Netlify	vercel.com · netlify.com — deploy on git push, free tier
GitLab repo	GitHub	github.com — free public + private repos

Section 2

Skill-by-Skill Practice

Standing Instructions

Project Context File

How to practice it now

Create CLAUDE.md in the root of any project you're currently working on
Write: what the project is, what stack it uses, how to run it, and any conventions you care about
Test it: start a fresh Claude Code session and ask "what do you know about this project?"

Delegation Contracts

How to practice it now

Before your next AI prompt, write the user story first — As a [user], I want [feature], so that [outcome]
Add 2–3 Given/When/Then acceptance criteria
Only then ask Claude to build it — and verify against the criteria when it's done
Try this on one real work task this week

Encoding Your Judgment

Skills

How to practice it now

Identify one thing you've asked Claude to do more than twice
Save it as .claude/commands/my-skill.md with the full instructions
Invoke it with /my-skill in the next session
Refine it after each use until it's reliable

Automated Quality Gate

Automated Tests from Acceptance Criteria

How to practice it now

Take a feature you've already built (work or personal project)
Write the Given/When/Then acceptance criteria for it after the fact
Ask Claude: "Generate unit tests for this feature based on these acceptance criteria: [paste criteria]"
Run them — fix until green

Section 3 · Your First Real-World Project

Start Here: The 1-Day Delegation Sprint

Pick something bounded — a script, a report generator, a simple internal tool. Then run the full Intermediate loop:

Set up context

Create CLAUDE.md. Write down what you're building and why.

Decompose

Break the project into 3–5 story-sized pieces. Write a user story for the first one.

Build with delegation contracts

Hand the first story to Claude. Verify against your acceptance criteria.

Capture a skill

After the first feature, save the pattern as a reusable skill.

Add tests

Ask Claude to generate tests from your acceptance criteria. Get them green.

Deploy

Push to GitHub. Connect to Vercel. Ship it.

Go parallel

Write stories 2 and 3. Delegate them in separate sessions simultaneously.

Section 4

Where to Go Next

Deepen This Track

Build the habits that compound

Revisit any session where you felt least confident
The hardest skill to build is delegation contracts — write one for every task for 30 days
Facilitator feedback doc: ask your facilitator for the link

Level Up to Advanced

The Delegator → Director track

The Advanced track builds directly on this one
New skills: evaluation harnesses, golden datasets, structured logging, quality gates, the autonomy slider
You're ready when delegation contracts feel natural and you want to scale across parallel workstreams

The concepts on this page work with any AI coding assistant — Claude Code, Cursor, GitHub Copilot, Windsurf. The tool changes. The delegation mindset doesn't.