$ cat cohort.md
Ship Production-Quality Code With Claude Code. Not Slop.
Claude Code writes your PR in 30 seconds. You spend 45 minutes reviewing it.
You find three things you'd do differently. You fix two. You ship the third because you're tired.
Next week you're debugging that third thing.
The tests pass — but they test what Claude wrote, not what you need. The PR is too big to review. The context fills up and you don't know why.
The tool is fast. But you're not saving time. You're just spending it somewhere else.
$ cat --story
The Problem Isn't You. It's the Workflow.
Most people use Claude Code out of the box. Default settings. No guardrails. Then they spend their day cleaning up after it.
I've used Claude Code every day for six months. I've trained 100+ engineers. I write the Claude Code Camp newsletter where I test the internals and publish what I find.
Same pattern everywhere. Fast output, slow cleanup.
When you know how the tool works — the loop, the context window, the costs — you stop cleaning up and start steering. That's what this course teaches.
Not for beginners. You should already use Claude Code and want to get better at it.
$ ./course --curriculum
What You'll Learn
Under the Hood
What's in your context window before you type anything.
- →Dissect the system prompt: the 27-31K tokens loaded before you type a word
- →How prompt caching works, what breaks it, and why a 1-line CLAUDE.md change can 10x your cost
- →CLAUDE.md vs .claude/rules/ vs .claude/settings.json: what goes where and why
- →Extended thinking: effort levels, when each fires, the real cost of 'high' on Opus
- →Read your own token usage: what a session actually costs you
A default CLAUDE.md burns ~2,100 tokens every turn. A lean one burns 280. You'll know the difference — and you'll cut your per-session cost by picking the right model and effort level instead of guessing.
Building Your System
Stop copy-pasting configs. Build something that works for your codebase.
- →gstack vs superpowers vs custom: what each costs in tokens per session
- →Write a skill from scratch that injects live data and self-corrects
- →Hooks: block dangerous commands, enforce lint, run tests before commit
- →Permissions: what --dangerously-skip-permissions actually does
- →Set up a feedback loop: tests + pre-commit + review so slop doesn't ship
- →Custom slash commands and /loop for recurring tasks
CLAUDE.md rules get followed ~70% of the time. Hooks enforce at 100%. You'll have both — plus a feedback loop that catches slop before it leaves your machine.
Building Beyond + Show & Tell
Claude Code is an agent loop. Now build your own.
- →The tool use loop: the exact cycle Claude Code runs on every turn
- →Build a working agent with the Claude Agent SDK in ~200 lines
- →When to use Claude Code vs Agent SDK vs raw API calls
- →Show & tell: demo what you built during the course
You'll have a working agent you built yourself. And you'll demo what you shipped during the course.
$ diff --before-after
Before and After This Course
OUTPUT QUALITY
STEERING
CONFIGURATION
CODE REVIEW
TESTING
MENTAL MODEL
$ cat reviews.txt
What Engineers Say
Engineers from these companies have joined:
"I learnt 3-4 different shortcuts/workflows that I can immediately add to my workflow to optimize, and it'll save me considerable time each day."
Student
Senior Engineer
"A huge injection of knowledge and motivation. I've carried it into my company — sharing ideas with colleagues and management."
Student
Senior Engineer
"Hugely productive. It solidified things I've been doing and gave me a list of things to explore. I've been doing this quite some time, so it's refreshing to get your perspective."
Student
Senior Engineer
"I've been using superpowers for the past week thanks to you, and I loved it. Most of the time I just want to accept its suggestions."
Student
Software Engineer
$ cat --includes
What's Included
- →3 live sessions (60 min each) with Q&A
- →Pre-reads and recorded demos, available immediately
- →Exercises to apply on your own codebase
- →Full session recordings so you can rewatch or catch up
- →Private Discord community with other cohort members
- →Optional office hours for 1:1 questions
$ claude-code-camp --instructor
Who's Running This?
Abhishek Ray
Trained 100+ engineers on Claude Code
Founder, Opslane (YC S24) · Ex-Robinhood Engineer
I write the Claude Code Camp newsletter. I run experiments on Claude Code internals and write up what I find.
Open source:
→claude-md-templates — CLAUDE.md starter templates
→claude-meter — real-time Claude Code session analytics
→Opslane — parallel Claude sessions for engineering teams
$ cat --decide
Two Ways This Goes
You keep reviewing 45-minute PRs. You keep shipping code you can't explain. You keep burning tokens on a context window you've never looked at.
Or you spend three Fridays learning how the tool actually works. And the output gets better from here.
30 seats. 3 sessions. Starts April 17.
$ ./checkout --cohort
Get Your Seat
3 sessions. Fridays at 9 AM PT. April 17, 24, May 1. Pre-reads available immediately. 30 seats.