wrangler-login

name	wrangler-login
description	Non-interactive wrangler OAuth login using playwriter to automate the Cloudflare consent flow in the user's browser. Use when wrangler auth is expired ("Failed to fetch auth token: 400 Bad Request") or when `wrangler whoami` fails. Delegate to a subagent Task to save context.

wrangler-login

Automate wrangler login when the OAuth token is expired and the agent cannot run interactive commands. Uses playwriter to click through the Cloudflare consent page in the user's existing Chrome session.

Prerequisites

The user must already be logged in to Cloudflare in their browser before you attempt this flow. The Cloudflare OAuth consent page is protected by a Turnstile captcha that agents cannot bypass. If the user is not logged in, Cloudflare will show a login page with Turnstile, and the automation will fail.

If Cloudflare is not already logged in: stop immediately and ask the user to log in to https://dash.cloudflare.com in their browser first, then retry.

When to use

• wrangler whoami fails or returns an auth error
• Deploy commands fail with "Failed to fetch auth token: 400 Bad Request"
• Any wrangler command says "set a CLOUDFLARE_API_TOKEN environment variable"

How to run

Delegate to a general subagent Task so the playwriter back-and-forth doesn't eat your main context. Pass this as the prompt:

Automate wrangler login via playwriter. Follow these steps exactly:

IMPORTANT: This only works if the user is already logged in to Cloudflare
in their browser. The consent page has a Turnstile captcha that agents
cannot solve. If at any point you see a Turnstile challenge, a login form,
or a CAPTCHA, STOP and tell the user to log in to https://dash.cloudflare.com
in their browser first, then retry.

1. Start wrangler login in a tuistory background session:
   bunx tuistory launch "pnpm exec wrangler login" -s wrangler-login
   bunx tuistory -s wrangler-login wait "/https:\/\/dash\.cloudflare\.com/i" --timeout 15000
   bunx tuistory read -s wrangler-login

2. Copy the full OAuth URL from tuistory output.

3. Load the playwriter skill (run `playwriter skill`), then open a
   playwriter session and navigate to that URL:
   playwriter session new
   playwriter -s <id> -e '
   state.page = context.pages().find(p => p.url() === "about:blank") ?? (await context.newPage());
   await state.page.goto("<OAUTH_URL>", { waitUntil: "domcontentloaded" });
   await waitForPageLoad({ page: state.page, timeout: 8000 });
   console.log("URL:", state.page.url());
   await snapshot({ page: state.page, showDiffSinceLastCall: false }).then(console.log);
   '

4. The consent page shows "Wrangler wants to access your account".
   Click through in order:
   a. Select the account button (e.g. "[e-mail]'s Account")
   b. Click "Review permissions"
   c. Wait 2s for permissions list to load
   d. Click the Authorize button: [data-test-id="oauth-consent-form-allow-button"]

5. After authorize, the page redirects to
   https://welcome.developers.workers.dev/wrangler-oauth-consent-granted
   and wrangler prints "Successfully logged in."

6. Verify: pnpm exec wrangler whoami

7. Clean up:
   bunx tuistory -s wrangler-login close
   playwriter -s <id> -e 'await state.page.close()'

Return the wrangler whoami output to confirm success.

Gotchas

• wrangler login must be running before you navigate to the OAuth URL. It starts a local HTTP server on port 8976 that receives the callback. If you navigate first, the redirect to localhost will fail.

• The consent page requires selecting an account before "Review permissions" becomes clickable. If you click Review without selecting, it shows "Select at least one account".

• The Authorize button uses data-test-id, not role. Use state.page.locator('[data-test-id="oauth-consent-form-allow-button"]') instead of getByRole.

• If the user has multiple Cloudflare accounts, the consent page shows multiple account buttons. Pick the one that owns the Workers/D1 resources.

• The OAuth URL contains a code_challenge that is unique per wrangler login invocation. Never reuse a URL from a previous run.