neuron-explorer
Steering-first neuron explorer for Qwen3.5-4B: push every MLP neuron, keep every rollout, browse the results
翻譯準備中
個人 GitHub 帳號下公開儲存庫的擁有者可以登入並驗證所有權,自行管理介紹,無需管理員核准。
登入並產生驗證檔案,在 30 分鐘內將其提交至此儲存庫的預設分支。其他分支或分叉儲存庫中的檔案無效。然後在下方檢查檔案,即可無需管理員核准直接管理專案。請保留檔案以維持管理權限。
如需登入,將前往登入頁面。登入後請回到此處,再次點擊所選的驗證按鈕。
在專案資料夾中執行 /nomorevibe verify,即可驗證所有權並自行管理介紹。
一分鐘完成驗證Template for sites that do things, not just show them: a React app and its API on one Cloudflare Worker, set up with Claude Code. AI calls with hidden keys, forms saved to D1, CI and deploys on the free plan.
翻譯準備中
取自儲存庫 README
正在載入完整 README。 以下為已儲存的摘錄。
For sites that do things, not just show them. The moment your site needs to keep a secret or remember something - an AI feature, a form that saves, a page only your clients see - it needs code on a server. This GitHub template gives you one: a React app and its /api/* back end on a single Cloudflare Worker, set up with Claude Code and running on the free plan. Most sites don't need it. Pick by what your site has to do: [README 설치·사용 부분 발췌] Quick start You need Node.js 24 (pinned in .nvmrc), a free Cloudflare account, a GitHub account and pnpm. No pnpm yet? Run `np檢視完整 README ↗
For sites that do things, not just show them. The moment your site needs to keep a secret or
remember something - an AI feature, a form that saves, a page only your clients see - it needs code on
a server. This GitHub template gives you one: a React app and its /api/* back end on a single
Cloudflare Worker, set up with Claude Code and running on the free
plan.
Site: https://henkisdabro.github.io/cloudflare-workers-react-boilerplate/
Most sites don't need it. Pick by what your site has to do:
| Your site... | Use |
|---|---|
| shows things - portfolio, brochure, landing page, docs | Plain files on Workers static assets, or Astro. Not this. |
| lives on search - blog, content-led shop, anything where ranking and link previews come first | A server-rendered framework: Astro, React Router or TanStack Start. Not this. |
| does things - calls an AI model with your key, saves what people send, shows each client their own data | This template. |
It is built for one person or a small team shipping an app with Claude Code: a paid AI tool, a booking or waitlist form backed by a database, a client portal, an internal dashboard. Workers run per request, so it is not the place for a long-running Node.js server.
pnpm create cloudflare?Cloudflare's own starter gives you the same React and Worker setup in one command - use it if you are happy wiring up the rest yourself. This template adds the parts a solo builder otherwise spends their first week on:
/start takes you through Cloudflare credentials, GitHub secrets, a custom
domain and the first deploy.main only after lint, types, tests and
build pass.AGENTS.md, slash commands and guides written for Claude
Code.Sign-in is not built in. The KV sessions example covers the cookie side; for a client portal, Cloudflare Access in front of the app is the quickest route.
@vitejs/plugin-react and the
Cloudflare Vite plugin, so pnpm dev
runs your Worker in the real workerd runtime alongside hot reload.worker/index.ts) with GET /api/health, a catch-all /api/* route and
security headers on API responses. Everything else is served as static assets with SPA fallback..github/workflows/deploy.yml: lint, type-check, test and build on every pull
request and push to main, then deploy from main.minimumReleaseAge in
pnpm-workspace.yaml, so a version published in the last three days refuses to install.You need Node.js 24 (pinned in .nvmrc), a free
Cloudflare account, a GitHub account and pnpm. No pnpm yet? Run
npx get-pnpm once and open a new terminal; from then on pnpm switches to the version pinned in
packageManager by itself.
Click Use this template on GitHub.
Clone, install and run - swap in your repository's URL and folder name, and leave it running:
git clone https://github.com/you/your-app.git && cd your-app && pnpm install && pnpm dev
The app is at http://localhost:5173, the API at /api/health.
In a second terminal, open the project in Claude Code the way you usually do - cd your-app && claude,
the desktop app or your editor's extension.
Type /start into Claude Code, or follow Manual setup.
/start renames the project from its placeholder name, replaces the template's AGENTS.md with a
project version, collects your Cloudflare account ID and API token, adds the GitHub secrets, optionally
sets up @claude on GitHub and a custom domain, and deploys.
name in package.json and wrangler.jsonc both ship as init; the
Wrangler name becomes your Worker's name and its workers.dev address.docs/PROJECT-AGENTS-TEMPLATE.md's starter block over AGENTS.md
and fill it in. The template's own AGENTS.md describes the template, not your project.CLOUDFLARE_API_TOKEN and CLOUDFLARE_ACCOUNT_ID.CHANGELOG.md and the promotional site
(docs/index.html, docs/404.html, docs/.nojekyll, docs/css/, docs/js/, docs/README.md).
LICENSE is the template's MIT licence in its author's name - put your own name in it, or delete it.main. The workflow deploys and prints the URL. If it ran before the secrets existed,
re-run it from the Actions tab..
├── src/ # React app (App.tsx, main.tsx, test/setup.ts)
├── worker/index.ts # Worker: /api/* routes and security headers
├── public/ # Static files copied as-is
├── examples/ # Copy-in reference code: ai/ and database/
├── docs/ # Beginner guides, plus the GitHub Pages site (index.html)
├── PRPs/ # Plans written by /generate-prp
├── .claude/ # Slash commands, /start branches, templates and settings
├── .github/workflows/ # deploy.yml (CI and deploy)
├── wrangler.jsonc # Worker config: assets, observability, commented binding examples
├── worker-configuration.d.ts # Generated by `pnpm cf-typegen` - commit it
├── vite.config.ts # React + Cloudflare plugins
├── vitest.config.ts
├── tsconfig.*.json # Separate projects for app, worker and node tooling
├── AGENTS.md # Agent instructions (CLAUDE.md imports it)
├── .mcp.json # Cloudflare API MCP server for Claude Code
├── CHANGELOG.md # Template release notes - /start deletes it
└── LICENSE # MIT - /start asks whether to keep it
| Command | What it does |
|---|---|
pnpm dev | Vite dev server with the Worker running in workerd |
pnpm build | Type-check all projects (tsc -b) and build |
pnpm preview | Build, then serve the production output locally |
pnpm lint | ESLint |
pnpm test | Vitest once (pnpm test:watch to watch, pnpm test:coverage for a report) |
pnpm run deploy | Build, then wrangler deploy (pnpm deploy is a pnpm built-in) |
pnpm cf-typegen | Regenerate worker-configuration.d.ts after any binding change |
Before pushing, run pnpm lint && pnpm test && pnpm build. CI runs the same three, plus
wrangler types --check.
Wrangler is a dev dependency, so call it through pnpm: pnpm exec wrangler tail,
pnpm exec wrangler secret put NAME, pnpm exec wrangler d1 migrations apply DB --remote.
Pick one way to deploy to production. They can all coexist with local pnpm run deploy.
.github/workflows/deploy.yml has two jobs:
main: frozen-lockfile install,
wrangler types --check, lint, test, build.main only, using
cloudflare/wrangler-action. It is skipped in the
template repository itself and runs once your generated repository is no longer a template.It needs two repository secrets: CLOUDFLARE_API_TOKEN and CLOUDFLARE_ACCOUNT_ID. Node and pnpm
versions come from .nvmrc and packageManager, so CI matches your machine.
Workers Builds builds and deploys from Cloudflare when you push, posts preview URLs on pull requests and needs no GitHub secrets. To switch:
deploy job from .github/workflows/deploy.yml. Keep verify so pull requests are still
checked.name in wrangler.jsonc.pnpm run build, and add a PNPM_VERSION build variable matching
packageManager - the build image ships an older pnpm by default.See limits and pricing for build minutes on each plan.
pnpm exec wrangler login # once
pnpm run deploy
Alternatively, put CLOUDFLARE_API_TOKEN and CLOUDFLARE_ACCOUNT_ID in .env (see .env.example).
wrangler.jsonc carries commented D1, KV, R2 and Workers AI bindings. KV, D1 and Queues bindings may
leave out their ID, and R2 its bucket name: wrangler deploy creates the resource and writes the ID
back to wrangler.jsonc
(automatic provisioning).
Run that first deploy from your machine and commit the result - a CI deploy creates the resource but
writes the ID only into the runner's throwaway checkout. The API token needs edit permission for each
product it provisions.
Run pnpm cf-typegen after any binding change, or CI's wrangler types --check fails.
The template runs on the Workers Free plan: 100,000 requests a day and 10 ms of CPU per request, with
free tiers for D1, KV and R2. Workers Logs are on by default (observability in wrangler.jsonc) and
move to Cloudflare Observability pricing from 2026-12-01. Check
Workers pricing and
limits for current figures, and
docs/cloudflare-products-explained.md for a plain-English
overview.
cp .dev.vars.example .dev.vars and fill it in. Values appear on the Worker's env
during pnpm dev. .dev.vars is gitignored.pnpm exec wrangler secret put ANTHROPIC_API_KEY.| Command | Use it to |
|---|---|
/start | Set up a new project from the template - run this first |
/generate-prp <feature> | Research a feature and write a plan to PRPs/ |
/execute-prp <file> | Implement a plan and run its validation gates |
/add-ai-feature | Add Claude API, Workers AI or AI Gateway features |
/setup-database | Add D1 or KV, with migrations and types |
/add-binding | Add R2, Queues, Durable Objects and other bindings |
/setup-sandbox | Add the Cloudflare Sandbox SDK for isolated code execution |
Details are in .claude/README.md. .mcp.json connects the
Cloudflare API MCP server,
which lets Claude Code inspect and manage resources in your Cloudflare account.
To have Claude answer @claude mentions on pull requests and issues, see
docs/GITHUB_ACTIONS_CLAUDE.md.
| Guide | Covers |
|---|---|
| CLOUDFLARE_WORKERS.md | Bindings, Worker patterns and operations |
| AI_INTEGRATION.md | Claude API, Workers AI and AI Gateway |
| SANDBOX.md | Sandbox SDK |
| examples/README.md | Copying the AI and database examples into your app |
| docs/GITHUB_ACTIONS_CLAUDE.md | @claude in GitHub pull requests and issues |
| docs/choosing-deployment.md | Deployment options in plain English |
| docs/cloudflare-products-explained.md | Workers, KV, D1 and R2 for beginners |
| docs/glossary.md | Terms used in these docs |
| docs/CONVENTIONS.md | Spelling and punctuation conventions |
Issues and pull requests are welcome on henkisdabro/cloudflare-workers-react-boilerplate. Read AGENTS.md first: every file here is copied into every generated project, so changes must hold true in projects you have never seen.
MIT.