cloudflare/workers-sdk wrangler@4.86.0

on GitHub

Minor Changes

• #13605 ea943ff Thanks @danielgek! - Add namespace support to wrangler ai-search commands

  All wrangler ai-search instance commands (create, list, get, update, delete, stats, search) now accept a --namespace (or -n) flag to target a specific AI Search namespace. When the flag is omitted, commands default to the default namespace that Cloudflare automatically provisions for every account.

  wrangler ai-search list now displays a namespace column, and wrangler ai-search create offers an interactive picker for existing namespaces (with an option to create a new one) when --namespace is not supplied in an interactive session.

  A new wrangler ai-search namespace subcommand group is also introduced, with list, create, get, update, and delete subcommands for managing namespaces directly.

  wrangler ai-search list --namespace blog
  wrangler ai-search create my-instance --namespace blog --type r2 --source my-bucket
  wrangler ai-search namespace create blog --description "Blog content"

• #13637 9eb9e69 Thanks @edmundhung! - Add --tunnel flag to wrangler dev for sharing your local dev server via Cloudflare Quick Tunnels

  You can now expose your local dev server publicly by passing --tunnel:

  wrangler dev --tunnel

  This starts a Cloudflare Quick Tunnel that gives you a random *.trycloudflare.com URL to share. The tunnel stops automatically when the dev session ends. Quick tunnels don't require a Cloudflare account or any configuration.

  A warning is shown when Server-Sent Events (SSE) responses are detected through the tunnel, since quick tunnels don't support SSE.

• #13661 0a5db08 Thanks @aspizu! - wrangler tail will now log stack traces. These stack traces already include resolved frames if you have chosen to upload sourcemaps.

• #13617 118027d Thanks @roerohan! - Force Flagship bindings to always use remote mode in local dev

  Flagship bindings now always access the remote Flagship service during local development, matching the behavior of AI bindings. Previously, Flagship supported both local and remote modes, but the local stub only returned default values, providing no real functionality and creating a dual source of truth for flag evaluations.

  The remote config field is retained for backward compatibility but only controls whether a warning is displayed. Setting remote: true suppresses the warning that Flagship bindings always access remote resources and may incur usage charges in local dev.

• #13254 e867ac2 Thanks @tgarg-cf! - Add wrangler queues consumer list subcommands for listing queue consumers

  Three new commands are available for listing consumers on a queue:

  • wrangler queues consumer list <queue-name> — lists all consumers (both worker and HTTP pull), grouped by type
  • wrangler queues consumer worker list <queue-name> — lists only worker consumers
  • wrangler queues consumer http list <queue-name> — lists only HTTP pull consumers

Patch Changes

• #13696 62e9f2a Thanks @dependabot! - Update dependencies of "miniflare", "wrangler"

  The following dependency versions have been updated:

  Dependency	From	To
  workerd	1.20260424.1	1.20260426.1

• #13576 2dc6175 Thanks @MattieTK! - Restore telemetry tracking for common CLI flags that were unintentionally dropped during sanitisation

  When argument sanitisation was introduced, only explicitly allow-listed args had their values included in telemetry. The allow list was very conservative, which meant common boolean flags like --remote, --json, --dry-run, --force, and many others were no longer being captured in sanitizedArgs despite previously being tracked. Boolean flags are inherently safe (values are only true/false), so these have now been added back to the global allow list. A small number of fixed-choice args (--local-protocol, --upstream-protocol, --containers-rollout) have also been added with their known value sets.

• #13649 ae8eae3 Thanks @petebacondarwin! - Fix service binding and tail consumer props being dropped between workers in different local dev instances

  When a service binding or tail consumer configured with props targeted a worker running in a separate wrangler dev instance (via the dev registry), the props were silently dropped and the remote entrypoint saw an empty ctx.props. Props are now forwarded correctly across the dev registry boundary, matching the behavior users get when all workers run in a single instance.

  // wrangler.json
  {
    "services": [
      {
        "binding": "AUTH",
        "service": "auth-worker", // may be in a separate `wrangler dev` process
        "entrypoint": "SessionEntry",
        "props": { "tenant": "acme" }
      }
    ]
  }

  The target worker's SessionEntry entrypoint now correctly receives { tenant: "acme" } on ctx.props regardless of which local dev instance it runs in.

• #13662 f2e2241 Thanks @petebacondarwin! - Fix three resource leaks in unstable_startWorker teardown that could prevent Node from exiting cleanly after worker.dispose().

  • The esbuild context created by bundleWorker is now disposed when the initial build fails. Previously a failing initial build (e.g. an unresolvable entrypoint, or a worker started with an invalid config via setConfig) left the esbuild child process running for the lifetime of the parent Node process.
  • runBuild's cleanup function now awaits the in-flight build before running the bundler's stop handler. Previously teardown could return before esbuild.BuildContext.dispose() had been called, so the esbuild watcher kept the event loop alive after dispose had resolved.
  • BundlerController.teardown() now runs the esbuild cleanup before removing the bundler's temporary directory, and aborts the in-flight bundle build so it cannot emit stale bundleStart/bundleComplete events after teardown. Previously the tmpdir was removed first, which in race with an in-flight rebuild produced confusing "Could not resolve .wrangler/tmp/bundle-XXXX/middleware-loader.entry.ts" errors during dispose.

• #13674 4f6ed93 Thanks @petebacondarwin! - Stop emitting a misleading [wrangler:error] Docker build exited with code: <n> log when the user aborts an in-progress container image build (for example by pressing the r rebuild hotkey while the previous build is still running).

  The abort-detection branch in the local and multi-worker runtime controllers was matching the wrong error message — it checked for "Build exited with code: 1", but the error thrown by the docker build helper is actually "Docker build exited with code: <n>", and the exit code after a process-group SIGINT/SIGKILL is typically 130/137/143, not 1. As a result, every legitimate user-initiated rebuild abort produced a spurious error event and [wrangler:error] log line. The check now matches the real error message prefix and ignores any non-zero exit code from the aborted build, so a user-requested rebuild while another build is in progress is silent.

• #13667 ed2f4ec Thanks @emily-shen! - fix: Preserve auth in remote proxy session data to avoid unnecessary session restarts

  maybeStartOrUpdateRemoteProxySession was not including auth in its return value, so on subsequent calls preExistingRemoteProxySessionData.auth was always undefined. This caused the auth comparison to always detect a change, disposing and recreating the remote proxy session on every reload even when auth had not changed.

• #13695 92bb8a5 Thanks @alexanderniebuhr! - wrangler types --check no longer throws when the types file was generated with an explicit boolean flag. Previously, yargs would parse such flags as actual booleans rather than strings, causing an internal parse error.

• #13662 f2e2241 Thanks @petebacondarwin! - Fix the wrangler tail command leaking a signal-exit listener after the tail has been cleanly closed.

  The tail command registered both a tail.on("close", exit) listener and a process-level onExit(exit) handler, but never removed the latter after exit() had run. In long-lived CLI processes this is harmless — the handler eventually runs once on shutdown — but in unit tests that repeatedly invoke wrangler tail, every invocation accumulates a handler that fires during test-runner shutdown. Those late invocations call deleteTail() after the test's auth mocks have been torn down, producing spurious "Not logged in" unhandled rejections which fail the Linux CI runs.

  The handler is now removed as soon as exit() runs, and exit() is guarded against re-entry so it is idempotent if both the WebSocket close event and a real signal fire for the same session.

• #13187 fcc491a Thanks @dario-piotrowicz! - Recognize Hydrogen as a known unsupported framework in autoconfig

  Previously, Hydrogen projects were incorrectly identified as React Router (since Hydrogen uses React Router under the hood), leading to a confusing autoconfig experience. Hydrogen is now recognized as a distinct unsupported framework, so users see a clear message that Hydrogen is not yet supported instead of being guided through React Router configuration.

• #13628 e6c437a Thanks @emily-shen! - fix: prioritise CLOUDFLARE_ACCOUNT_ID over a cached account id for all Pages commands

  Previously, some Pages commands (pages deploy, pages deployment list/delete/tail, pages download config, pages secret) used a cached account id over the CLOUDFLARE_ACCOUNT_ID environment variable. The pages project commands already correctly prioritised CLOUDFLARE_ACCOUNT_ID.

• Updated dependencies [21b87b2, 62e9f2a, 033d6ec, ae8eae3, ef24ff2, 6d27479, 118027d]:

  • miniflare@4.20260426.0