The Operator Kit for Claude Code
Files and rules for running Claude Code as a full time, multi agent operator. Built for one person at a terminal, not a team dashboard.
Independent product. Not affiliated with or endorsed by Anthropic.
The problem
A Claude Code setup drifts. You add a rule for one bad incident, forget to remove it, and six months later nobody can say which rule still matters.
External lanes die quietly, too. In our own setup, Codex, Cursor, and Grok once sat unauthenticated for days and nothing said so. Work fell back to the most expensive model, and the bill was the first warning.
Fan outs make it worse. Spawn ten workers with no model set, and each one inherits your priciest model by default. Nobody writes the handoff either, so the next session starts cold.
What's inside
Plain files: markdown skills and rules, four agent wrappers, five shell scripts, one hooks JSON, one Workflow example, and a README with a 30 minute quick start.
Skills
- orchestrator-boot. The boot ritual: read doctrine, snapshot live state, write a state brief, then act. Ships with
boot.sh, one command that pulls usage, sessions, repos, and lane health into a single file. - observe-repair. Closed loop verification. Judge a fresh observation of real state, never the source code or a clean build.
- unslop. Cuts AI tells out of any writing.
- brand-voice-template. The method for measuring a writing voice from a corpus, not inventing one. Ships empty. You fill in your own tables.
- review-checklist. The must fix, should fix, ok pattern, with an evidence rule behind every finding.
- skill-creator-notes. What makes a skill trigger reliably and what makes one fail silently.
Rules
- autonomy-tiers.md. A T0 to T4 escalation framework: what you can do without asking, and what needs a human first.
- delegation-and-roles.md. Orchestrator versus worker roles, cheap lane routing, and the rule to set the model explicitly on every worker.
- output-style.md. Answer first, plain sentences, no padding.
- LESSONS.md. The traps: verify before you alarm, numbers point to their source, never trust a self report, a dead lane stays silent, and more.
Agents, briefs, scripts, hooks, workflows
Four wrapper subagents (codex-lane, cursor-lane, grok-lane, notion-lane): the first three run on haiku and forward one task to an external CLI, returning the output verbatim; the last is a sonnet data lane for a workspace tool. Brief, handoff, and state templates, tied together by RITUAL.md. Five shell scripts for usage, lane health, browser routing, and spawning a worker terminal. One hooks JSON with three hooks: block a remote push, print usage on stop, warn on em dashes. One Workflow script example with the model set explicitly on every lane.
The screenshot
How it fits together
- Boot. Read doctrine, read the last handoff, run the boot script, write a fresh state file.
- Brief. Write the task down: exact files, constraints, acceptance criteria, how to verify.
- Delegate. Route mundane work to a cheap model. Route external CLI work through a wrapper subagent, model set explicitly.
- Verify on disk. Diff the repo or read the file yourself. A worker's report is a claim, not a fact.
- Handoff. Write down what shipped, what proves it, and the traps you hit, so the next session doesn't start cold.
Who it's not for
If you want a chat wrapper that answers questions and stops there, this isn't it. The kit assumes you're running Claude Code from a terminal and delegating real work to it.
If you don't use the terminal, skip it. Every piece here is a file you read, edit, or run from a shell.
Get the kit
$49, one time, no subscription. You get every file in the manifest: the skills, the rules, the four agent wrappers, the brief templates, the scripts, the hooks JSON, and the Workflow example, plus the README.
The kit doesn't include a Claude subscription, API credits, or the external CLIs it wraps. You need Claude Code already. Codex, Cursor, and Grok are optional.
Checkout runs through Stripe. After payment you land on the download page and get the zip, no waiting on email.
Refunds: 14 days from purchase, no questions asked. Email support and you get the refund.
Want to see the files before you pay? The hooks guide shows the full hooks JSON, and the subagents guide shows a complete wrapper file, both straight from the kit.
Version 0.1, released September 2026. The changelog ships in the zip.
FAQ
Does this include Claude?
No. You need your own Claude Code install. The kit is files that run inside it.
Which Claude Code version does it need?
The current release. The kit is plain files. Claude Code changes often, so the changelog records what we adjust, and updates come through the same download link.
Do I need Codex, Cursor, or Grok?
No. The four wrapper subagents are optional.
What's the refund policy?
14 days from purchase, no questions asked. Email support and you get the refund.
How do updates work?
Free updates for every release through v1.x, delivered through the same download link.
What's the license?
Single buyer, personal use, on any number of your own machines. No resale or redistribution of the kit itself.
Is this affiliated with Anthropic?
No. Independent product. Not affiliated with or endorsed by Anthropic.
Guides
Support: wilsoncreative.business@gmail.com.
Independent product. Not affiliated with or endorsed by Anthropic.