Most people use like 15% of Claude Code. I built a free tool that teaches the rest by baking the features into your own plans.

Ok so I've basically been living in Claude Code for months, and one day it hit me that I was probably using like 15% of it. Plan mode, subagents, skills, hooks, MCP, /security-review, auto-memory... all right there, and I just never touched them. Not because they're hard — I just never knew *when* I was supposed to reach for them. So I built a thing to fix that for myself. It's called Adeptly. Free, open source, MIT, runs on your machine. You tell it what you want to build, it writes you a proper plan, and here's the part I actually care about: it drops the right Claude Code features straight into the plan, in the spots where they make sense, with a quick note on what each one does and why it's there. So you kind of learn Claude Code just by reading your own plans instead of digging through docs. That was the whole itch I was scratching. npx adeptly It opens on localhost, reads your plans from docs/plans/, and uses YOUR local claude CLI for anything AI. No API key, no backend, no telemetry, no sign-up. Your code never leaves your machine — if you already trust Claude Code, this doesn't add anything new to trust. New in v0.5, which I shipped today: the Crew. Up till now it would just *tell* you the workflow to run. Now it can actually run it. You get a little pipeline of roles — Architect → Approval Gate → Builder → Medic (runs your build/tests and patches them up if they break) → Reviewer → Security → Pilot, which opens the PR. Each one is just a headless claude turn in your own repo. So all those features it kept nudging you about... they actually get used. I was a bit nervous about the "let it run in my repo" part, so: - It's dry-run by default. Simulates the whole thing, touches literally nothing. Run it as many times as you want just to see how it flows. - Going live is deliberately annoying to do by accident: the plan has to be approved AND you have to start it with ADEPTLY_LIVE=1. Both. Otherwise it just refuses. For context, it got around 1k downloads on a pretty quiet first launch, but honestly v0.5 is the first version I'm not embarrassed to share. Repo's public now too, so if you want to poke holes in it or send a PR, please do — genuinely want the feedback, good or bad. Happy to answer anything about how it works under the hood.