The Locked In Method for Claude Code
Claude can't say "done" until your tests pass, can't touch your secrets, and uses less of your limit. Copy one folder into ~/.claude and you're set up in two minutes.
rm -rf ~, force-push to main and reading .env are blocked before they run./handoff → /clear → /pickup keeps context small without losing your place.Try it. A simulation of what the kit's hooks and workflows do in a real session.
What's in the box
All set up for you: 43 files that drop into ~/.claude. Here's every one, grouped by what it does for you.
Expert playbooks Claude pulls in by itself when the task matches, plus six workflows you start with one word.
you type itloads by itself
Every skill that changes code ends the same way: run your real checks, show the output, and fix until they pass.
Specialists that work in their own context and report back in a paragraph, so your main session stays small.
Rules that run every time, with no exceptions and no tokens spent.
Short project memory for your stack: the commands, gotchas and checks Claude can't guess.
The Locked In Method: the five-step loop, habits that make your limit last, and 10 prompts that work.
See the method ↓The Locked In Method
Anthropic's best-practices guide puts this habit first. Claude Code's creator, Boris Cherny, says it improves quality 2–3×. The kit turns it into five steps, and every step is a file you install.
Claude interviews you about the hard parts, then writes a spec with one command that proves the feature works.
/lock-inSPEC.mdYou agree on the plan before any file changes. Skip this when the change fits in one sentence.
Shift+TabarchitectClaude runs the check after each step. The Stop hook won't let it finish while the check fails.
verify-gate.pyauto-format.pyYou get real test output, then a fresh reviewer checks the diff against the spec. Only gaps that matter get fixed.
/second-opinioncode-reviewerWrite down where you are, clear the context, and pick up in a fresh session. You keep your progress, and the context stays small.
/handoff/clear/pickupLook inside
Plain Markdown and Python with nothing compiled or hidden. Files marked with a dot show a real excerpt.
How the pieces fit
Claude Code already supports all of these. Most people never set them up because it takes weeks to get them right.
| Piece | Count | Runs when | Lives in |
|---|---|---|---|
| Skills | 12 | Load on their own when the task matches, e.g. "is this ready to deploy?" loads ship-it | ~/.claude/skills/ |
| Subagents | 6 | Claude hands off a review, search or plan to its own context | ~/.claude/agents/ |
| Workflows | 6 | You type /lock-in, /handoff, /pickup, /mvp, /second-opinion, /cleanup | ~/.claude/skills/ |
| Hooks | 5 | Every time, no exceptions: before a command runs, after an edit, and when Claude tries to finish | .claude/settings.json |
| CLAUDE.md templates | 8 | Loaded at the start of every session in that project | ./CLAUDE.md |
| Playbook | 1 | You read it once, in about ten minutes | PLAYBOOK.md |
Unzip, run one block, restart Claude Code. Type / to see the workflows and /agents to see the subagents.
The playbook
A large context is resent with every message you send. The playbook shows how to keep it small without losing your place.
Plan Mode gets you an agreed plan before any file changes. One good plan costs less than three rewrites.
Write down the state, wipe the context, and pick up from a short file instead of a huge history.
Jump back to an earlier message instead of arguing the model out of a wrong direction.
Reference files by path so Claude reads only what it needs, once.
The researcher subagent reads thirty files in its own context and hands back one paragraph.
It loads into every session. Every rule has to be worth its cost on every message.
If it doesn't save you time, ask within 14 days for a full refund.
Questions
/handoff → /clear → /pickup loop, the Haiku-powered researcher and the context-saver skill keep your context small, and a smaller context uses less of your limit on every message. The verify gate also cuts the back-and-forth of "that didn't work, try again."/lock-in, /mvp and the hooks give you the habits experienced engineers use without thinking, and the verify gate catches broken code you wouldn't know how to spot.