Source: cloudflare/skills · Maintainer: Cloudflare · License: Apache-2.0 · Checked: 24 October 2026
What is the wrangler skill?
The wrangler skill is Cloudflare's official Agent Skill for running and troubleshooting Wrangler CLI commands in an existing Workers project. It makes your coding agent check the project's installed Wrangler version and the current Cloudflare docs before it writes a command, edits config, deploys or touches secrets.
Its SKILL.md description reads: "Run or troubleshoot Wrangler CLI commands in projects that have a wrangler.jsonc, wrangler.json, or wrangler.toml file, including local development, Previews, deployment, and resource management. For new projects and projects without a Wrangler configuration file, use the cf CLI instead."
The premise is stated plainly: "CLI flags and configuration fields change; do not rely on memorized examples." As of 24 October 2026, its skills.sh page listed 127.6K installs (snapshot, 24 October 2026).
What does it do?
The folder holds a single file, SKILL.md. It is a procedure in five parts rather than a command cheat sheet.
1. Inspect the project. The agent finds the package manager, the installed Wrangler version, package scripts, the framework and the Wrangler config. It runs Wrangler through the project's scripts so the local version is used, and the skill says "do not silently upgrade Wrangler to match current docs." It also establishes the target account, Worker, environment and resource before running anything that changes them, and whether a data operation is local or remote.
2. Retrieve what the task needs. It prefers the Cloudflare MCP docs tool when available, or fetches the linked page directly. A routing table maps each task to a source: project-local wrangler --help and the Wrangler command reference for flags, the installed wrangler/config-schema.json and the configuration reference for bindings, plus pages for environments, secrets, TypeScript types, local development, permissions, Previews and migrations from Pages or Vercel. If retrieval fails, the agent should "state the gap" instead of guessing syntax.
3. Apply the change. The rules here are about avoiding surprises:
| Situation | What the skill tells the agent |
|---|---|
| New config | Prefer wrangler.jsonc and set the compatibility date to today |
| Adding bindings per environment | Check inheritance, since some fields must be set for each environment |
| Dashboard edits exist | Reconcile them first, because Wrangler can overwrite dashboard variables and routes |
| Local dev with remote bindings | Check which bindings are real before testing writes |
| Remote commands | Identify the logged-in member or API token and use the narrowest scope |
| Secret values | Keep them out of arguments, source and logs |
wrangler secret put or secret delete |
Treat it as a deployment, or use wrangler versions secret to stage it |
| Rollback | Check rollback limits: connected resources and their data are not rolled back |
It also notes that the wrangler login OAuth flow "does not support granular authorization," so an account-owned API token is the route when you need narrow access, and the agent should never ask you to paste that token into chat.
4. Work with Workers Previews. The skill says Previews require project-local Wrangler 4.135.0 or later, and that a newer global install does not count. It tells the agent to mirror how you already manage production (Wrangler, dashboard or Terraform), to warn you that Preview URLs are public unless access controls are set, and to pass the same --env value to every Preview command when targeting a Wrangler environment. Cloudflare's Previews get started page confirms the 4.135.0 minimum, and the Previews docs cover the feature itself.
5. Validate. After config or binding changes in TypeScript, the agent regenerates types with wrangler types and runs your existing checks. For deploy changes it uses wrangler deploy --dry-run where supported, with the caveat that a dry run "does not prove remote resources or runtime behavior work." It ends by reporting what changed, the target environment, checks run and any gaps.
How do you install it?
Cloudflare's README lists several routes. Pick one per agent.
Just this skill, with the skills CLI. This is the command shown on skills.sh:
npx skills add https://github.com/cloudflare/skills --skill wrangler
Add -g to install it for all projects. For Claude Code, the folder lands in .claude/skills/wrangler/ in your project, or ~/.claude/skills/wrangler/ globally, the standard Claude Code skill locations.
Claude Code plugin (every Cloudflare skill plus the MCP server). From the README:
/plugin marketplace add cloudflare/skills
/plugin install cloudflare@cloudflare
The plugin bundles Cloudflare's main remote MCP server, which gives the agent "the Cloudflare API and current developer documentation." That is the docs tool the skill prefers for retrieval.
Codex plugin.
codex plugin marketplace add cloudflare/skills
codex plugin add cloudflare@cloudflare
The README says to start a new Codex session afterward.
Cursor, VS Code and manual copy. Cursor installs from its marketplace or as a remote rule pointing at cloudflare/skills. VS Code with GitHub Copilot installs from source after you enable chat.plugins.enabled. You can also clone the repo and copy the folder into your agent's skills directory. For agent-specific details, see Codex skills and Cursor skills.
How do you use it?
Ask for Wrangler work in plain words inside a project that has a Wrangler config. Example prompts (suggestions, not quotes from the repo):
Add an R2 bucket binding called UPLOADS to staging only, then regenerate the types.Deploy this Worker to production. Do a dry run first and tell me which account and environment you're targeting.Set the STRIPE_KEY secret on staging without putting the value in a command or log.Set up Workers Previews for pull requests, using the same CI we use for production.Our last deploy broke checkout. Roll back and tell me what a rollback won't undo.
You need Wrangler in the project (the skill adds it locally if a config exists but Wrangler is missing) and network access if you want the agent to read Cloudflare docs. The skill holds no keys; remote commands use your own Wrangler login or API token.
Tips and limits
- It hands off new projects. In a project with no Wrangler config, or one with
cloudflare.config.ts, the skill tells the agent to use the Cloudflare CLI (cf) and "Do not add Wrangler to those projects." Installcf-clitoo if you start new projects, or read Cloudflare's cf for coding agents page. - It leans on retrieval. The SKILL.md links to the docs instead of copying flags. An agent without web or MCP access gets the procedure but has to rely on local
--helpoutput. - Secrets deploy immediately. The skill flags that
wrangler secret putcreates and deploys a version. If you need staging, ask for thewrangler versions secretworkflow. - No scripts. The folder is one Markdown file and runs nothing by itself.
- Audits. On 24 October 2026, skills.sh showed Pass from Gen Agent Trust Hub, Socket and Snyk.
- License. Apache-2.0 lets you copy and change it; keep the license and notices when you redistribute.
Save it in your Kipt library
In Kipt, open Skills and paste the skill's GitHub folder link into the import box. For a repo with many skills, like this one, Kipt shows a list, so you can take durable-objects or workers-best-practices in the same pass. You can also drop the folder or a .zip. Your skills live in your own private GitHub repo and install into Claude Code or Codex with one command:
npx skills add https://github.com/<you>/<repo>.git --all -g
Kipt is not affiliated with Cloudflare.
Similar skills
- Cloudflare Workers skill:
workers-best-practices, for writing and reviewing the Worker code that Wrangler deploys. - Durable Objects skill: from the same repo, for stateful objects, SQLite storage and alarms.
- Cloudflare Agents SDK skill: for stateful AI agents that run on Workers.
- Web performance skill: Cloudflare's Core Web Vitals audit for the pages you deploy.