wrangler

Patch Changes

• Allow wrangler dev to start module Workers without a default export

  Previously, wrangler dev failed to start module Workers that exported only named entrypoints, such as RPC services, unless an unused default export was added. These Workers now start normally in local development without that workaround.

• Fetch the remote config diff from the configured compliance region

  When a Worker was last deployed from the dashboard, wrangler deploy and wrangler versions upload download its remote configuration to compare with the local one. These requests ignored the compliance_region field in the Wrangler configuration file and only honoured the CLOUDFLARE_COMPLIANCE_REGION environment variable. A FedRAMP High Worker configured through the config file would send them to api.cloudflare.com instead of api.fed.cloudflare.com, and the deploy would fail with authentication errors. They now use the same compliance region as the rest of the deploy.

• Fail with an error instead of hanging when local dev requests can't work under Bun

  Bun's fetch() ignores the undici dispatcher that Miniflare routes requests through (oven-sh/bun#39247). Under Bun, createTestHarness()'s server.fetch() and server.getWorker()'s fetch(), email() and scheduled() now throw a UserError, requests to the dev server URL get a 503 naming the cause, and starting remote bindings throws a UserError. Each of these used to hang with no error. server.getWorker().getEnv() and getExport() keep working under Bun.

• Stop --strict deploys and version uploads from aborting when the API omits optional binding fields

  Previously, wrangler deploy --strict and wrangler versions upload --strict could report a destructive change for an unchanged configuration, for example an R2 bucket in the default jurisdiction, because a field the API left out was compared as removed. Missing and unset optional fields are now treated as equal, so these commands proceed when the local configuration matches the deployed Worker.

Patch Changes

• Keep remote dev sessions alive through transient network failures

  wrangler dev --remote and getPlatformProxy() refresh their preview token before it expires, but a single failed refresh — for example while the network was briefly down — stopped refreshing for good, leaving the session broken until restart. A failed refresh now retries every minute until it succeeds. Rebuilds also no longer push the refresh past the session's expiry.

• Fix wrangler rollback reporting success after the changed secrets confirmation is declined

  Previously, declining the "The following secrets have changed…" confirmation printed "Aborting rollback..." and then also printed a success message and the target version as the "Current Version ID", even though no deployment was made and the original version was still live. Wrangler now stops after aborting, so the output matches what is actually deployed.

• Hide the wrangler dev hotkey bar when Turborepo does not forward input

Minor Changes

• Include source maps by default in wrangler build --experimental-cf-build-output

  Build Output now matches wrangler deploy and wrangler versions upload, which upload source maps by default, so Workers built for tools like cf get stack traces that point at the original source. Source maps that can't be resolved are skipped rather than failing the build. Set upload_source_maps = false to opt out.

Patch Changes

• Support local D1 queries through the experimental d1_binding_jsrpc compatibility flag

  Local D1 bindings now implement the RPC query backend used by D1Database.query() and prepared-statement methods when the flag is enabled. Queries reuse local SQLite execution, including transactional batches, execution metadata, and session bookmarks, while bindings without the flag continue to use HTTP. Update the bundled workerd runtime to 1.20261012.1 and align the default compatibility date with that release.

Minor Changes

• Upload source maps by default for wrangler deploy, wrangler versions upload and wrangler preview

  When upload_source_maps isn't set in config or via --upload-source-maps, Wrangler now generates and uploads source maps so stack traces in logs and wrangler tail point at your original source without extra setup. In this default mode, source maps that can't be resolved (for example inline data: maps or missing .map files from a custom build) are skipped rather than failing the upload. Setting upload_source_maps = true keeps the existing strict behaviour, and upload_source_maps = false (or --no-upload-source-maps for deploy and versions upload) opts out.

  wrangler build --experimental-cf-build-output is unchanged and still only includes source maps when upload_source_maps = true.

• Stop wrangler deploy, wrangler versions upload and wrangler preview in projects that build for the Cloudflare CLI (cf)

  If your project builds through cf, deploying it with Wrangler reads a different configuration, so it could update the wrong Worker or fail without explaining why. These commands now stop before uploading anything, name the .cloudflare/output directory they found, and tell you to run cf deploy instead. If that directory is just left over from an earlier build, deleting it restores the Wrangler workflow.

  Ordinary Wrangler projects are unaffected, and --dry-run still works on deploy and versions upload. This is a new guard for the experimental Build Output Specification rather than a fix, and it gives three stable commands a failure mode they did not have before, so it is a minor.

Patch Changes

• Clarify the Worker not found error during CI deploy validation

  Wrangler now reports that the configured Worker could not be found in the account instead of incorrectly describing the failure as a name mismatch.

• Ignore .wrangler writes in the dev assets watcher

  wrangler dev reloaded in a loop when assets.directory was the project root, because the assets watcher treated wrangler's own writes under .wrangler as asset changes and dropped in-flight requests. The watcher now ignores the project's .wrangler directory and dev.persist, except when that path is the assets directory.

• Fix optional Docker lookups printing Docker CLI errors during successful builds

  cf build looks for Container images left behind by earlier builds so it can remove them. This lookup is best effort, but when Docker could not be reached it still printed errors such as Cannot connect to the Docker daemon, even when the build succeeded and no image needed Docker. Docker commands whose output Wrangler reads no longer forward the Docker CLI's error output to the terminal. If one of these commands fails in a way that stops the command, the error message still includes the Docker CLI's output.

• Time out the OAuth token request instead of waiting indefinitely

  The request that exchanges an authorization code or refresh token for an access token had no timeout, so a stalled connection (for example behind an HTTP proxy) could leave wrangler login hanging silently for many minutes after the browser step had succeeded. The request now gives up after 30 seconds and reports that the Cloudflare auth server could not be reached.

• Preserve binding options when provisioning resources through a redirected config

  Wrangler now writes provisioned resource identifiers to both the redirected and original config files without copying other binding options between them. This prevents generated relative paths such as D1 migrations_dir from replacing user-authored paths during Vite deployments.

• Honor X-Forwarded-Proto for incoming development requests behind TLS-terminating proxies

  Miniflare now uses the original client protocol when constructing URLs for network requests. This also prevents Wrangler from downgrading HTTPS URLs in response headers such as Location and WWW-Authenticate when a tunnel forwards requests over HTTP. Explicit programmatic request URLs and configured upstream origins remain authoritative.

Minor Changes

• Support temporary event accounts in R2 and Containers commands

  wrangler r2 and wrangler containers commands now accept the hidden --temporary flag, so accounts created for an event can manage buckets, objects and containers directly. Every command that supports --temporary now also accepts a hidden --event-code flag, so the first command a participant runs can create the event account:

  wrangler r2 bucket create my-bucket --temporary --event-code <code>

  R2 and Containers are only available on event accounts. R2 custom domains, Sippy, external container registries and wrangler cloudchamber commands still require a logged-in account.

Patch Changes

• Honor Retry-After directives during static asset uploads

  Static asset uploads now pause retries and pending uploads until the latest outstanding deadline requested by the API. A per-request limiter also keeps gateway retries at the reduced concurrency after the pause ends, preventing a deployment from immediately overloading a constrained asset service again.

• Document that addresses catch-all entries must use the zone apex

  A *@domain entry in addresses must use the zone apex, such as *@example.com. The zone catch-all also receives mail for every subdomain in the zone that has no literal rule.

• Keep Preview secrets when deploying to an existing Preview

  Secrets added to a Preview with wrangler preview secret put or wrangler preview secret bulk were lost the next time wrangler preview ran, because each new deployment was created from the Wrangler config, --var and --secrets-file values only.

  wrangler preview now carries over the secrets of the Preview's latest deployment. A value passed in this deployment (--secrets-file, --var or a previews binding with the same name) still replaces the existing secret, and wrangler preview secret delete removes one.

• Stop the update check from recommending deprecated versions

  Previously, the "update available" notice shown by wrangler and @cloudflare/vite-plugin always pointed at whichever version was tagged latest on npm, even after that version had been deprecated for shipping a bug. Deprecated versions are now never recommended: if the latest release has been deprecated, the newest non-deprecated stable release below it is suggested instead, or nothing at all if you are already on it.

  The check now reads the npm registry directly instead of going through the update-check package, which discarded the deprecation information. The on-disk cache location and one-hour refresh interval are unchanged.

• Ship using and await using declarations to the runtime as written, for smaller Worker bundles

  Workers and Pages Functions that use explicit resource management no longer carry about 1 KB of bundled helper code to emulate it. workerd supports using and await using natively at every compatibility date, so wrangler deploy, wrangler versions upload and Pages Functions builds now leave these declarations untouched.

• Apply each action's own condition in wrangler r2 bucket lifecycle add

  When a lifecycle rule was added with both expiration and Infrequent Access transition flags, both actions read their condition from the same list of flags, which checked --expire-days first. --expire-days 365 --ia-transition-days 30 therefore transitioned objects after 365 days instead of 30, and --expire-date 2027-01-01 --ia-transition-days 30 deleted objects after 30 days instead of on the given date. Expiration now only reads --expire-days/--expire-date, and transition only reads --ia-transition-days/--ia-transition-date.

Minor Changes

• Add --source-namespace and --source-repo-name to wrangler queues subscription create for the artifacts.repo source

  The Event Subscriptions API requires source.namespace and source.repo_name for artifacts.repo subscriptions, but Wrangler had no way to pass them, so --source artifacts.repo always failed with a validation error. Both flags are now required for this source, and wrangler queues subscription get shows the subscription's resource as <namespace>/<repo-name>.

• Add assets.base_path support to Workers Assets

  Serve an asset directory from a public URL prefix without changing its on-disk layout:

  {
    "assets": {
      "directory": "./public",
      "base_path": "/docs"
    }
  }

  Wrangler, preview, Miniflare, and generated build configuration preserve the explicitly selected value, while the Asset Worker normalizes it and strips the prefix only for asset lookup. Requests passed to a user Worker, request-facing headers, and redirects retain the public path. Relative pathname inputs are interpreted as root-relative prefixes, URL-shaped values are rejected, and omitting the option preserves existing root-path behavior.

  Authored _headers and _redirects rules continue to match full public paths. In particular, both the source and destination of an authored 200 asset rewrite must include the configured public prefix; Asset Worker-generated redirects are prefixed automatically.

• Add a --experimental-mode instant option to wrangler kv namespace create

  This lets entitled accounts create Workers KV Instant namespaces while the feature is in private beta.

• Add email-protected Quick Tunnels to wrangler dev

  Pass one or more --tunnel-allowed-mail flags to require email authentication when exposing a local development server through a Quick Tunnel. Each value can be an exact email address or a domain pattern.

• Support deleting secrets with wrangler versions secret bulk

  Set a secret's value to null in JSON input to remove it from the new Worker version. Bulk output now distinguishes between created and deleted secrets, so retrying wrangler secret bulk with wrangler versions secret bulk preserves requested deletions. Deploy the new version with wrangler versions deploy to apply the changes to production traffic.

Patch Changes

• Improve guidance for conflicting Wrangler configuration files

  When user and generated deploy configurations are found under different base paths, Wrangler now identifies the expected deploy configuration location, suggests how to resolve the conflict, and links to the relevant documentation.

• Include default module rules in generated Worker types

  wrangler types now declares the built-in Text, Data, and WebAssembly module patterns even when they are not repeated in the Wrangler configuration, keeping generated types aligned with deployment behavior.

  Service-worker declaration files are emitted as global scripts so that the generated wildcard module types are visible to imports.

  Directory-specific rules retain their scope when TypeScript can represent it; ambiguous relative imports use a union of the possible deployed module types.

  When generating combined types for named environments, each environment's effective rules are resolved independently and differing import types are represented as unions.

• Fix r2 object put and r2 bulk put storing a different key in local mode

  Keys that are not URL-safe were mangled on the way into local storage.

  • A key with a space or a non-ASCII character was stored under its percent-encoded name, so a later r2 object get for that key reported that the key does not exist.

  • Two keys that differ only after a # collapsed into a single object, and the second upload replaced the first.

  • A key with a % that is not a valid escape failed outright with "Invalid URL string.", and one with a valid escape, such as %41.txt, was stored as A.txt.

  Spaces, non-ASCII characters, # and % now survive the trip into local storage. Objects already in local state are left where they are.

• Show a useful error when wrangler secret bulk hits an undeployed latest version

  wrangler secret put already explained this case (API error 10215). secret bulk just dumped the raw API response, which for 10214 talks about logpush and tail_consumers even though you were only uploading secrets.

  Both commands now point at wrangler versions secret … instead.

• Print temporary account notices to stderr

  The terms notice, the proof-of-work message, and the "Temporary account ready" claim details printed by --temporary now go to stderr instead of stdout. Previously they corrupted command output on stdout, such as the JSON from wrangler kv namespace list --temporary or the raw value from wrangler kv key get --temporary. Commands that lower the log level for --json, such as wrangler d1 execute --json --temporary, also hid the claim URL; it is now shown unless logging is disabled with WRANGLER_LOG=none.

  Scripts that read the claim URL from stdout should read stderr instead.

Minor Changes

• Allow "us" as a jurisdiction for Container applications

  Container placement constraints now accept constraints.jurisdiction: "us" in Wrangler and typed Cloudflare configuration. This makes the US jurisdiction available alongside "eu" and "fedramp".

Patch Changes

• Fix wrangler containers list to report live instances

  The LIVE INSTANCES column now reports each application's active runtime instances instead of its configured instance count, matching the Cloudflare dashboard. JSON output continues to expose the configured count through the existing instances field.

• Accept Durable Object application IDs in Containers commands

  wrangler containers instances and wrangler containers delete now accept the 32-character hexadecimal application IDs returned for Durable Object-backed applications, in addition to legacy dashed UUIDs.

• Retry transient API failures in wrangler workflows instances list and wrangler workflows instances describe

  Previously, a single temporary 5xx response or dropped connection made these read-only commands exit with an error, even though the next request would have succeeded. They now use Wrangler's existing bounded API retry handling. The read that resolves --id latest is retried too, which also benefits the other wrangler workflows instances commands that accept latest; the mutating requests they make afterwards are not retried. Persistent failures are still reported after the retries are exhausted, and under --json any retry notices are written to stderr so stdout stays valid JSON.

Minor Changes

• Support the new Workflows createBatch() API in local development

  Local Workflows bindings now accept object-form batches that create instances from a count or a list of instance options. The result includes handles for created instances and indexed per-instance errors, matching the runtime API while preserving the deprecated array form.

• Add --allowed-mail to the experimental wrangler tunnel quick-start command

  The option forwards exact email addresses, comma-separated lists, and wildcard domains to cloudflared. It can be specified more than once to combine multiple recipient rules.

  Email-protected tunnels require cloudflared 2026.9.2 or later. Wrangler checks the selected binary before starting the tunnel and reports an upgrade error when it is incompatible.

Patch Changes

• Mark wrangler artifacts commands as open beta

  Artifacts has entered open beta, so the wrangler artifacts commands no longer display a "private beta" label in help output and warnings.

• Keep colons in wrangler tail --header filter values

  wrangler tail --header splits its argument into a header name and an optional value at the colon. It split at every colon and kept only the first two parts, so a value containing a colon was cut short: --header "Origin:https://app.example.com" filtered on https. The value now includes everything after the first colon, so URLs, ports and IPv6 addresses are sent to the tail filter intact.

Minor Changes

• Add native support for the Analytics SQL binding

  Declare the zero-configuration binding in wrangler.json with "analytics": { "binding": "ANALYTICS" }. Wrangler uploads the analytics binding type and proxies it to the remote service during local development, so wrangler dev can call the binding without unsafe.bindings.

• Graduate SQL, Catalog, and Pipelines under wrangler basin out of beta to stable

  Basin SQL is now available under wrangler basin sql, Basin Catalog operations are available under wrangler basin catalog, and Pipelines operations are available under wrangler basin pipelines. These commands are now stable, while the previous wrangler r2 sql, wrangler r2 bucket catalog, and wrangler pipelines command paths remain available as hidden compatibility aliases.

  The Basin SQL authentication environment variable is now WRANGLER_BASIN_SQL_AUTH_TOKEN. Update any existing WRANGLER_R2_SQL_AUTH_TOKEN configuration to use the new name. The fallback to CLOUDFLARE_API_TOKEN remains available.

• Add beta K2 producer bindings for existing streams

  Configure a stream created through Wrangler, the Dashboard, or the API in wrangler.json:

  {
    "k2": [
      {
        "binding": "ORDERS",
        "stream": "0123456789abcdef0123456789abcdef"
      }
    ]
  }

  The binding supports env.ORDERS.send([{ content: new TextEncoder().encode("order"), headers: { event: "order.created" } }]). Batches use either all ArrayBuffer or all Uint8Array content. Check the returned success value, handle rejected RPC promises, and retry only when the returned error explicitly allows it. Generated environment types describe this producer contract without requiring a separate application dependency.

  K2 requires an enabled account. Deployment credentials need Worker deployment and K2 configuration-read access. Default Wrangler logins now request the K2 OAuth scopes; existing OAuth users should run wrangler login again to grant the new permissions. Development always uses a real K2 stream and may incur usage charges; no local simulator is provided. The remote setting can be omitted, remote: true suppresses the usage warning, and remote: false is rejected. Consumption is not part of this Worker binding.

• Add beta K2 stream management commands

  Use wrangler k2 streams create order_events, wrangler k2 streams list, wrangler k2 streams get <stream-id>, and wrangler k2 streams delete <stream-id> to manage K2 streams. Creation enables Worker bindings but not HTTP ingestion by default, matching the dashboard. Pass --http-enabled to enable authenticated HTTP ingestion and print its endpoint. Creation prints the stream ID and a binding configuration with a YOUR_BINDING_NAME placeholder for the Worker's variable name, but does not edit the configuration file automatically.

  All four commands support --json. Deletion requires confirmation, or --force/-y to skip it; use --force --json for JSON deletion output. Creation also accepts retention, HTTP authentication, Worker-input, and CORS options; listing supports pagination and a name filter. Default Wrangler logins now request k2.read and k2.write; existing OAuth users should run wrangler login again, or use a custom API token granting K2 Config Write. The account must be enabled for K2.

Minor Changes

• Add --tty (-t) flag to wrangler containers ssh to force pseudo-terminal allocation

  OpenSSH only allocates a pseudo-terminal when no remote command is given, so interactive commands such as wrangler containers ssh <ID> -- bash previously ran without a prompt or line editing. Pass --tty to force one:

  wrangler containers ssh <ID> --tty -- bash

• Support SSH settings for Durable Object-managed Containers in the configuration API

  defineContainer now accepts ssh and authorizedKeys with schedulingPolicy: "durable-object", matching the ssh and authorized_keys fields that Wrangler already supports for these Containers. Previously the schema rejected them, so they could not be set from cloudflare.config.ts.

  defineContainer({
    name: "sandbox",
    schedulingPolicy: "durable-object",
    ssh: { enabled: true },
    authorizedKeys: [{ name: "laptop", publicKey: "ssh-ed25519 AAAA..." }],
  });

Patch Changes

• Fix wrangler dev remote bindings for workers.dev subdomains protected by Access

  Running wrangler dev with remote bindings on an unpublished worker protected by Access (e.g. using a wildcard on your workers.dev domain) previously failed with a redirect loop. Wrangler now correctly authenticates remote bindings with Access in this situation.

• Upgrade the bundled capnweb implementation to 0.12.0

  This updates the RPC implementation shipped in Miniflare and remote-binding proxy workers to the latest capnweb release.

• Resolve the affected Undici dependency in new Wrangler and Vite plugin installs

  Undici 7.29.1 fixes GHSA-3wwx-pv8p-q78v. Update the shared dependency catalog and matching types used by Miniflare and Wrangler so downstream installs can resolve the patched runtime without an application-level override. A published release is still required for consumers; this changeset does not alter already published package metadata.

• Fix custom-domain-only deploys failing for API tokens without Zone Workers Routes read permission

  When workers_dev was disabled and routes contained only entries with custom_domain: true, every deploy after the first one fetched /zones/:zoneId/workers/routes to check for route conflicts, even though custom domains are not zone Workers Routes. Tokens scoped to Workers Scripts edit plus custom domains - without Zone > Workers Routes > Read - failed with "No access to the specified resource" after the Worker version had already been uploaded. The conflict check now only covers non-custom-domain routes; custom domain conflicts continue to be reported by the custom domains changeset API.

• Report an unreachable auth server instead of an expired login when refreshing an OAuth token

  When the OAuth token endpoint could not be reached (for example a DNS failure or a connection timeout), the refresh failure was reported as "Your auth token has expired and could not be refreshed", with advice to run wrangler login; in an interactive terminal Wrangler also started a new browser login. A network failure says nothing about the stored refresh token, and a new login would need the same unreachable server. Wrangler now reports that the Cloudflare auth server could not be reached, leaves the stored credentials unchanged, and does not start a login, so the next run can refresh with the same token once the network is back.

Minor Changes

• Support Workflows declared in exports on ctx.exports in local development

  A Workflow declared in a Worker's exports is now available on ctx.exports in wrangler dev, the Vite plugin and the Vitest plugin, with the same API as a Workflow binding:

  const instance = await ctx.exports.MyWorkflow.create({
    params: { name: "World" },
  });

  ctx.exports and workflows bindings with the same Workflow name share their instances, including instances created before the Workflow was declared in exports. Two Workers can't export the same Workflow name, and a binding to an exported Workflow must refer to the Worker and class that export it. getPlatformProxy() ignores Workflows declared in exports, since it doesn't run the Worker's code.

  wrangler workflows commands run with --local also work with Workflows declared only in exports, without a workflows binding.

  In the Vitest plugin, introspectWorkflow() and introspectWorkflowInstance() still need a Workflow binding, and now explain how to add one when passed a Workflow from ctx.exports. Instances created through ctx.exports are introspected too. A workflows binding whose script_name is the Worker's own name now resolves to the Worker itself again.

Minor Changes

• Add Durable Objects code update strategies to Worker deployments

  Use --durable-objects-code-update-mode immediate with wrangler deploy, wrangler versions deploy, and wrangler rollback to update code without waiting for active instances to hibernate. Use --durable-objects-code-update-mode deferred 30s to set a maximum delay, or configure durable_objects.code_update_strategy with mode and max_delay. When unset, the strategy defaults to deferred with a 5-minute maximum delay; delays cannot exceed 24 hours and must use millisecond precision.

• Add Browser Run as an event source for Queue subscriptions

  You can now create Queue subscriptions with --source browserRun.

Patch Changes

• Show the stack and cause of failed proxied requests in wrangler dev debug logs

  When a request proxied to the local Worker fails, running with --log-level debug now shows the underlying error's stack and cause chain.

Minor Changes

• Support Containers in Worker Preview deployments with the Build Output.

Minor Changes

• Configure SSH for experimental Durable Object-managed Containers

  Set containers.ssh and containers.authorized_keys when using scheduling_policy: "durable_object". These are application-wide settings that follow the same rules as the existing Durable Object-managed Container settings: normal deployments create missing applications and update explicitly configured values, while omitted settings preserve the existing application configuration.

  // wrangler.jsonc
  {
    "containers": [
      {
        "name": "sandbox",
        "class_name": "Sandbox",
        "scheduling_policy": "durable_object",
        "ssh": { "enabled": true },
        "authorized_keys": [
          { "name": "laptop", "public_key": "ssh-ed25519 AAAA..." }
        ]
      }
    ]
  }

• Expose configured Cron Triggers to local development consumers

  Wrangler now passes the active environment's exact Cron Trigger expressions to Miniflare so Local Explorer can display them. Headless agent sessions also advertise the Local Explorer scheduled invocation API.

• Support UDP connect handlers in local development

  The experimental connect configuration now accepts protocol: "udp", with optional idle_timeout_ms and max_pending_bytes settings. UDP datagrams are delivered to the Worker's connect() handler using workerd's value-mode socket streams, and can be tested with Miniflare#dispatchConnect({ protocol: "udp" }).

• Support workflow entries in the exports configuration map

  A Worker can now declare the Workflows it defines in exports, keyed by the WorkflowEntrypoint class name:

  {
    "exports": {
      "MyWorkflow": {
        "type": "workflow",
        "name": "my-workflow",
        "limits": { "steps": 100 },
        "schedules": "0 * * * *"
      }
    }
  }

  A workflow export accepts the same settings as a workflows binding: limits, concurrency, schedules, and default_retention. wrangler deploy and wrangler versions upload send these entries to the upload API by name, and wrangler deploy and wrangler triggers deploy provision the Workflow with its settings, just as they do for workflows bindings owned by the Worker. A Workflow may be declared both as a binding and as an export, as long as both declarations use the same class and do not set the same setting to different values. A binding to another Worker's Workflow cannot share a name with an export. @cloudflare/config adds the matching exports.workflow() helper. Local development does not yet act on these entries.

Patch Changes

• Emulate the deterministic-ID uniqueness contract in the local Workflows binding

  The local Workflows binding now matches the documented production behavior for deterministic instance IDs: create({ id }) with an ID that already exists throws (instance.already_exists) and retains the existing instance, and createBatch() skips IDs that already exist or repeat within the batch, excluding them from the result instead of creating duplicate executions. Previously both paths silently created duplicates, so code relying on deterministic IDs for idempotency (for example a Queue consumer creating one workflow per message) appeared to work locally while double-executing workflow bodies.

• Show --jurisdiction in help for wrangler kv namespace create

  The option was supported but omitted from the command's help output. Users can now discover how to create KV namespaces in a specific jurisdiction.

• Update smol-toml to 1.9.0 to fix slow parsing of very large TOML files

  Parse time for TOML config files now grows linearly with their size, instead of with its square: a 40,000-line file that took 259 ms to parse now takes 17 ms, while typical wrangler.toml files parse in the same time as before. This addresses the GHSA-r4xh-jqrq-34v2 advisory against earlier versions of the parser.

  Some TOML syntax errors now point at the character that caused them. For example, a wrangler.toml containing INVALID "FILE is now reported as illegal character in key at the ", rather than incomplete key-value at the start of the line.

Minor Changes

• Add event-code support to temporary Worker deployments

  Use wrangler deploy --temporary --event-code <code> to provision an account for an event. Wrangler requires explicit server acknowledgement before caching the account and keeps the event code out of its cache and telemetry.

• Allow framework commands to produce Preview Build Output with the experimental config

  When cf previews deploy invokes a framework build command, Preview intent is now preserved. Function-based cloudflare.config.ts files receive isPreview: true, and generated Build Output is marked as a Preview build.

Patch Changes

• Standardize Zod validation error output

  Format validation errors with Zod's built-in prettifyError() helper so Miniflare, Wrangler, the Vite plugin, and the Vitest plugin show consistent messages and property paths.

Patch Changes

• Update smol-toml to 1.8.0

  This updates the bundled TOML parser that reads wrangler.toml to a version that addresses two advisories against 1.5.2: GHSA-7w5x-hrqm-74c2 (a value followed by a comment with no trailing newline, such as a=[1 #, put the parser in an infinite loop) and GHSA-v3rj-xjv7-4jmq (thousands of consecutive comment lines overflowed the stack). On the old version, wrangler deploy against a wrangler.toml ending in a=[1 # never returned; it now fails with Invalid TOML document: cannot find end of structure.

• Warn when wrangler preview returns only non-custom-domain URLs even though custom-domain Preview URLs are configured.

• Fix Preview output artifacts to always include the resolved parent Worker name

  Preview artifacts now use Wrangler's resolved Worker name instead of relying on the Preview API response to include it.

Patch Changes

• Fix wrangler types generating runtime headers with trailing whitespace

  Runtime type headers without compatibility flags now end at the compatibility date, keeping generated types reproducible when tools remove trailing whitespace.

• Improve Container image listing and deletion

  List all image pages using read-only credentials, validate tags before deletion, and report successful deletion when the garbage-collection request fails.

• Fix Container SSH connection setup and shutdown

  Prevent SSH connections from stalling during setup and ensure proxy processes exit when sessions close.

• Show valid sha256-prefixed tags in Container image listings

  Container image listings now distinguish valid OCI tags such as sha256-release from synthetic digest entries such as sha256:<digest>.

Patch Changes

• Improve wrangler preview onboarding guidance

  Wrangler now displays placeholder replacement guidance directly beneath the suggested Preview configuration instead of as a separate warning. JSON output continues to include the guidance in its structured onboarding messages.

• Read workers.dev URLs from the Worker resource during deployment

  Wrangler no longer requires account-level subdomain permission to display Worker and version-preview URLs. It now uses the Worker-scoped URL fields while preserving account-level registration for accounts without a workers.dev subdomain.

Minor Changes

• Identify experimental Build Output resource configs by filename and location

  The root remains config.json, Worker configs are now worker.config.json, and Container configs are now container.config.json. Resource configs no longer contain top-level type discriminators, while settings and build context are stored together in the root config.

• Define experimental Cloudflare configuration with a single default export

  Experimental cloudflare.config.ts files now define settings and resources together in a default-exported defineConfig() call. Add a Worker under worker, add Containers to the containers array, or omit both to provide settings only.

  import * as entrypoint from "./src/index.ts" with { type: "cf-worker" };
  
  export default defineConfig({
      accountId: "...",
      complianceRegion: "public",
      worker: {
          name: "my-worker",
          compatibilityDate: "2026-09-18",
          entrypoint,
      },
  });

• Add experimental --zone and --zone-id flags to wrangler deploy and wrangler triggers deploy to attach a zone to routes passed via --route

  Routes passed on the command line were always sent to the Cloudflare API as bare patterns. Zones with an SSL for SaaS entitlement reject such routes with error 10082 ("When using wildcard host ssl for saas entitlement you must specify the zone per route using zone_id or zone_name"), and until now the only way to set a zone was in the config file, which --route overrides.

  The new flags are experimental and must be enabled with --experimental-route-zones (alias --x-route-zones). Pass a single zone to apply it to all routes, or one zone per route in the same order as the --route flags:

  wrangler deploy --x-route-zones --route "app.example.com/*" --route "api.example.com/*" --zone example.com

  wrangler deploy --x-route-zones --route "a.example.com/*" --zone example.com --route "b.example.net/*" --zone example.net

  --zone sets zone_name and --zone-id sets zone_id on each route. The two flags cannot be combined, and passing more than one zone requires exactly one per --route. Routes without zone flags behave exactly as before.

• Remove the experimental Container image environment binding

  Durable Object-managed Containers now use ctx.container.images without Wrangler generating env.EXPERIMENTAL_CLOUDFLARE_CONTAINER_IMAGES. Update code using the experimental environment binding to read ctx.container.images and regenerate your Worker types.

  Version deployments identify managed applications from native named images, and --containers-rollout=none preserves native Container metadata. Containers without named images must first be provisioned with wrangler deploy; versions upload verifies that their applications already exist. The old binding is no longer read or reserved, including on previously uploaded versions. keep_vars retains existing variables as usual; redeploy without it to remove an existing experimental binding.

• Return structured configuration errors from wrangler preview --json

  When a Worker is missing its Preview configuration, JSON mode now returns an error, a suggested_config patch, and any associated onboarding messages without interactive output or terminal formatting. This changes the private-beta Preview command to make automated onboarding reliable.

• Add support for jurisdictions to Queues subcommands

Patch Changes

• Allow local Container images without exposed ports

  Wrangler and the Cloudflare Vite plugin no longer reject images that omit Docker EXPOSE metadata. Local Containers can run command-only workloads or serve traffic through workerd without declaring an unused image port.

• Fix cf builds for static projects that serve assets from the project root

  The experimental Build Output path now omits the reserved .cloudflare directory when the project root is used for static assets. This prevents recursive output copying in Wrangler while preserving the existing behaviour for other asset directories.

• Rebase absolute non-JavaScript module specifiers when preserve_file_names is enabled

  With preserve_file_names set, a non-JS module imported by an absolute path kept that path as its module name. The build machine's filesystem layout ended up inside the deployed Worker, and the module was never written to --outdir. A local dry run reported success while the upload failed server-side with error code 10021. Tooling that rewrites externals to absolute paths hits this, which is how it was found in @opennextjs/cloudflare with WASM imports.

  Absolute specifiers are now rebased to ./<basename>, which is what the hashed branch of the same code already does minus the hash prefix. Relative specifiers keep the behaviour they had.

Minor Changes

• Build Containers when emitting experimental Build Output

  Wrangler and the Cloudflare Vite plugin now build Dockerfile-backed Container images when experimental Build Output is enabled. Container configs are emitted under .cloudflare/output/v0/containers with local image references, while existing registry references pass through unchanged.

• Evaluate Flagship flags locally during development

  Flagship bindings now use the local Miniflare store by default in Wrangler and the Vite plugin, keeping development offline and isolated from production flags. Set remote: true on a binding to continue using its remote app.

  Use wrangler flagship flags pull <APP_ID> to seed the store from a remote app. Flag management commands also accept --local to read and update the local store directly.

• Pass Preview intent to defineWorker and upload its resolved configuration

  Preview builds now evaluate programmatic Worker configuration with ctx.isPreview set to true and record that intent in Build Output. The shared Preview uploader deploys the resolved bindings and settings while preserving configured Preview base values when it creates a Preview.

Patch Changes

• Fix duration calculation for running workflow instances, steps, and attempts in wrangler workflows instances describe

  wrangler workflows instances describe previously distorted the elapsed duration of in-progress instances, steps, and attempts across non-UTC timezones by stripping " GMT" from toUTCString(), causing new Date(...) to parse the timestamp in the local client timezone. The duration is now correctly computed against the current time.

Minor Changes

• Add support for configuring real-time Issues with observability.issues.enabled

  Wrangler now validates and uploads the Issues setting alongside the existing logs and traces observability options. The experimental configuration format supports the equivalent observability.issues.enabled option.

• Mark wrangler preview commands as open beta

  Wrangler now labels Preview commands as open beta in help output and command warnings, matching the feature's public availability.

• Support explicit named Container image selection in Wrangler local development

  Wrangler builds or pulls named images configured through Wrangler JSON or TOML and exposes their local tags through ctx.container.images. Pass one of those references to ctx.container.start({ image }) to select the image.

  This extends the experimental Durable Object-managed Containers interface. Named images are opt-in and do not become the Container's default image. A Container without a default image must supply an image or full Container snapshot when starting.

Minor Changes

• Add placement configuration for Previews

  You can now configure placement in the previews block. Preview-specific placement overrides the top-level placement configuration for Preview Defaults and deployments.

• Improve onboarding guidance for Previews (when previews block is missing from configuration file)

  When a local previews block is absent, Wrangler writes the Preview Base configuration to the local config file. When no Preview Base configuration exists, Wrangler prints a placeholder configuration derived from production bindings and warns against reusing production binding configuration.

Patch Changes

• Align Local Explorer Workflow instance status requests with production

  Local Explorer and Wrangler local mode now use the production-compatible status request field for pausing, resuming, restarting, and terminating Workflow instances. Direct Local Explorer API consumers must replace the previous action field with status.

  Successful Local Explorer status updates now return the production-compatible instance status and response timestamp instead of the local-only result.success acknowledgement.

• Send exports with Worker Preview deployments

  wrangler preview dropped the exports block from deployment requests. Durable Objects reached through ctx.exports had no Preview namespace, and cache settings for each entrypoint were lost too.

Minor Changes

• Categorise the positional path argument to wrangler deploy and wrangler versions upload in command telemetry

  Command telemetry now records a coarse category for the entry-point/assets positional (wrangler deploy <path>) under sanitizedArgs.path, so we can understand whether people pass a file, a directory, or a relational reference such as . or ../example. The possible values are file, directory, current-dir, parent-relative, and not-found, or null when no positional is provided. The raw path is never sent — only the category.

• Configure application-wide logs for experimental Durable Object-managed Containers

  Set containers.observability.enabled or containers.observability.logs.enabled when using scheduling_policy: "durable_object". Normal deployments create missing applications and update explicitly configured log settings without a Container rollout. Omitted settings preserve the application configuration; root Worker observability is not inherited for this policy.

  Version uploads may initialize missing applications but preserve existing settings. Deploying or rolling back Worker versions also preserves existing application settings, and --containers-rollout=none skips their updates.

• Support per-image build options for experimental Durable Object-managed Containers

  Set build_context and build_vars alongside dockerfile in a Container's named images entries. Context paths resolve relative to the Wrangler configuration file and default to the Dockerfile's directory. Build variables are passed as Docker build arguments. Entries using the same Dockerfile with different contexts or variables are built separately.

  {
    "containers": [
      {
        "class_name": "Sandbox",
        "scheduling_policy": "durable_object",
        "images": {
          "app": {
            "dockerfile": "./docker/Dockerfile",
            "build_context": ".",
            "build_vars": { "APP_ENV": "production" }
          }
        }
      }
    ]
  }

• Support AI Search bindings in Worker Previews

  wrangler preview now accepts ai_search and ai_search_namespaces entries in the previews block and includes them in Preview deployment bindings. This lets Workers that use AI Search instance or namespace bindings attach existing resources to Preview deployments, including preview-specific instance or namespace names.

  These bindings are non-inheritable: declare them explicitly under previews. They attach to existing AI Search resources; preview does not provision new isolated instances or namespaces.

• [private beta]: Add --secrets-file and --var flags to wrangler preview

  Like wrangler deploy and wrangler versions upload, wrangler preview now accepts a --secrets-file flag pointing to a JSON or .env format file, and --var KEY:VALUE pairs that are injected into the Preview deployment as plain text variables. CLI vars override same-named vars from the previews section of your config file, and secrets from the file take precedence over both:

  wrangler preview --secrets-file .env.preview --var API_URL:https://api.example.com

• Remove the gated Web Search binding and Wrangler command

  The unreleased search binding and its experimental command have been removed from Wrangler, Miniflare, and configuration APIs.

• Allow experimental Durable Object-managed Containers to link by name through exports

  Containers using scheduling_policy: "durable_object" can now specify name and link from exports.<Class>.container without repeating class_name. Deploy and version upload resolve that link for image preparation, Worker metadata, and Container application creation.

Patch Changes

• Sync Local Explorer endpoint lists across agent hints

  The Local Explorer endpoint list is now consistent across the three places it appears: the AGENTS.md template in create-cloudflare, the runtime agent hint in wrangler dev, and the Vite plugin agent hint. All three now include the observability/clear endpoint, use the canonical /cdn-cgi/local/explorer path, and have cross-reference comments pointing to each other.

• Fix per-query overrides for wrangler ai-search search

  --score-threshold, --max-num-results, --filter, and --reranking are now sent using the AI Search request schema, so the service applies them to searches instead of ignoring them.

• Surface the original error message, name and stack when the dev server reports an internal error

  Previously wrangler dev could exit with an empty ✘ [ERROR] log that gave no indication of what went wrong (e.g. Network connection lost., see #14641). These errors now include their original message, name and stack, so the failure is actually diagnosable.

• Treat 502, 503, and 504 as gateway errors during asset upload retries

  Pages and Workers asset uploads now retry more patiently when the Cloudflare API responds with a 502, 503 or 504 gateway error, reducing concurrency and waiting longer between attempts instead of failing the deploy quickly.

• Improve over-limit run_worker_first errors when duplicate rules are present

  The error now reports distinct and duplicate-entry counts and lists duplicated rules, making it clear when removing redundant entries can bring the configuration within the limit.

  Too many `run_worker_first` rules were provided; 105 rules provided (99 distinct, 6 duplicate entries) exceeds max of 100. Note: duplicate entries count towards the route limit. Ensure that no duplicate rules are present in your `run_worker_first` configuration.
  
  The duplicated rules found are:
  - "/rule/0"
  - "/rule/1"
  - "/rule/2"
  - "/rule/3"
  - "/rule/4"
  ...and 1 more duplicated rule.

• Replace execa with tinyexec for running subprocesses, shrinking the bundled Wrangler output.

• Preserve service-worker middleware error propagation with spec-compliant event dispatch

  Wrangler's synthetic service-worker events now propagate listener exceptions to middleware without changing the behavior of user-created EventTarget instances.

• Reduce the size of Wrangler's published package

  Exclude test-only, build-only, and obsolete template files from the npm package while retaining all runtime templates.

• Restore static asset upload concurrency after gateway errors

  Static asset uploads previously remained at concurrency one for the rest of the deployment after any 524 response, which could make large deployments exceed the upload session lifetime. Successful uploads now restore the session's original concurrency gradually while retaining gateway throttling. Requests that were already in flight when throttling began do not restore capacity, so a burst of stale successes cannot immediately undo backpressure.

Patch Changes

• Add a provisioning delay note when custom domain Preview URLs change

  Wrangler now explains that DNS and TLS certificate provisioning may continue after a deploy adds a custom domain or enables its Preview URLs. Stable redeploys don't repeat the note.

  This assumes that a request which matches the stored custom domain state doesn't restart provisioning. The client infers this from the API changeset and current domain record because this repository can't verify the backend behavior.

• Clarify production status labels for custom domain routes

  Wrangler now prefixes explicit custom domain production states with production: so they match Preview labels. The updated labels appear in deployed trigger output and WRANGLER_OUTPUT_FILE_PATH.

• Avoid replacement prompts for custom domains already on the Worker

  Wrangler now updates Preview settings without asking to replace a custom domain when that domain already belongs to the deployed Worker. It still asks before replacing domains attached to another Worker.

• Explain how to enable Preview URLs when a Preview deployment has none

  wrangler preview now shows URL shapes and configuration snippets for Workers.dev and custom domains. The custom domain snippet preserves every configured route, and the guidance distinguishes missing settings from disabled ones.

  This changes a private beta feature. The warning also makes clear that wrangler deploy publishes code from the current checkout.

Minor Changes

• Add Durable Object-managed Containers to top-level container configuration

  Wrangler now accepts scheduling_policy: "durable_object" in the top-level containers array and creates its namespace-backed application after the Worker upload resolves the Durable Object namespace ID. The namespace ID is also the application ID, so repeated deploys idempotently ensure the same application without name-based lookup, modification, or a Containers rollout.

  Durable Object-managed entries accept class_name, scheduling_policy, an optional name, and an optional named images map. Scheduler-only fields are rejected. Each image provides either a local dockerfile or a digest-pinned managed-registry image. Wrangler builds or resolves each image, waits while Cloudflare prepares it for the Containers runtime, and uploads the resulting references with the Worker version for access through ctx.container.images and env.EXPERIMENTAL_CLOUDFLARE_CONTAINER_IMAGES. Local development support for these entries is deferred to a follow-up.

  Existing scheduler-backed entries and Durable Object migrations continue to work unchanged.

  With --containers-rollout=none, existing Workers retain their deployed Container metadata and image binding even when local containers is omitted or empty; local scheduler edits are also ignored. The upload stops if the deployed versions cannot be recovered. Existing Workers for Platforms dispatch scripts reject this flag before upload because their API does not expose enough metadata to preserve Container associations safely. First deployments can still skip Container preparation and rollout. Without this flag, removing managed Containers, including by omitting containers entirely, clears the experimental image binding even with keep_vars.

  versions deploy validates the selected versions before changing traffic and creates their Durable Object-managed applications only after deployment succeeds. Both deploy and versions deploy report partial completion if application creation fails afterward, with instructions to retry the same command.

  EXPERIMENTAL_CLOUDFLARE_CONTAINER_IMAGES is a temporary, reserved Wrangler binding until native Container image metadata is available. Its class keys identify managed applications during versions deploy, including classes with empty image maps. User configuration cannot declare a binding with this name; existing versions that already use it are treated as Container configuration.

• Remove wrangler preview settings commands

  The private-beta wrangler preview settings and wrangler preview settings update commands are no longer available.

Patch Changes

• Fail wrangler versions upload early when a Worker has a pending Durable Object migration

  Wrangler now directs users to run wrangler deploy to apply the migration instead of sending a version upload request that the API will reject.

• Detect named-only module Worker entrypoints correctly

  Wrangler now distinguishes named-only module Workers from legacy Service Workers that happen to have named exports. A default export identifies a module Worker; otherwise, legacy addEventListener registration identifies Service Worker format.

• Correct Pages-to-Workers delegation analytics for forced and ineligible commands

  The legacy forced result counted every agent-driven Pages command using --force, including commands that could never have been delegated. Wrangler now emits eligible_forced only when --force prevents an otherwise eligible delegation, and records other agent commands as ineligible with a bounded reason and whether force was used.

• Prevent delayed internal errors from fetch-only remote bindings

  Fetch-only remote bindings such as D1 and R2 previously opened an unused WebSocket RPC session. RPC sessions are now created only when an RPC method is called.

• Add the missing transferred_classes migration to the config schema

  DurableObjectMigration described new_classes, new_sqlite_classes, renamed_classes and deleted_classes, but not transferred_classes. normalizeAndValidateConfig has always validated that key, and the deploy path forwards it to the API along with the rest of the step, so Transfer migrations worked — but config-schema.json is generated from the type, so an editor resolving $schema reported a valid, documented migration as an unknown key.

  Adding the field to the type puts it in the generated schema. No runtime change.

• Bump shell-quote to 1.9.0+ to pick up two disclosed advisories

  shell-quote@1.8.1 is affected by a ReDoS in parse() (CVE-2026-13311 / GHSA-395f-4hp3-45gv — an unauthenticated attacker who can feed a string into parse() can block the event loop for tens of seconds with plain space-separated input, no shell metacharacters required) and by an object-token escaping bug in quote() (CVE-2026-9277 / GHSA-w7jw-789q-3m8p), both fixed upstream in 1.9.0. Wrangler's parse() wrapper (src/utils/shell-quote.ts) is reachable from pages dev/init command-line parsing, so the ReDoS applies; the quote() call site only ever passes string arguments, so the object-token issue was not reachable here, but there is no reason to stay on a vulnerable range once a patch exists.

• Encode filenames in Pages HTML redirects

  Fix wrangler pages dev returning a 502 response when redirecting HTML paths containing Unicode characters. Keep reserved characters in filenames encoded in the redirect destination and preserve the request query string.

• Fix wrangler types --strict-vars=false emitting invalid TypeScript for an empty array var

  A var whose value was an empty array produced (), which is a syntax error. Because this lands in the generated worker-configuration.d.ts, it did not just break that one line — the whole file failed to parse, so no binding types resolved at all. An empty array now generates unknown.

• Use previews_base_config for Preview configuration

  Preview commands now read the Worker Previews Base configuration from the previews_base_config API field.

• Fix wrangler workflows instances describe crashing on dynamic retry delays

  The Workflows API serializes function retry delays as "[dynamic]". The describe command previously parsed that as a duration, produced an Invalid Date, and threw RangeError: Invalid time value before printing remaining steps. It now renders unknown (dynamic delay) and also tolerates attempts whose end timestamp is missing.

Minor Changes

• Add containers.observability support to wrangler deploy

  Wrangler now accepts container-specific observability settings via containers.observability, including application-level targeting fields for Containers. Root observability continues to work as a fallback when a container does not define its own observability settings.

  wrangler deploy now preserves legacy configuration.observability for existing container apps that still use rollout-based observability, while using top-level application observability for new or already-migrated apps.

  Existing application diffs are now normalized even when stored resource limits cannot be mapped to a named instance type. API-only metadata and equivalent managed-registry image names no longer appear as edits or affect whether deployment changes require a rollout.

• Delegate agent Pages project creation with a production branch to Workers

  When run by an AI agent, wrangler pages project create --production-branch <name> is now eligible for delegation to a Workers static-assets deploy. The production branch names the target that a Workers deploy would publish to, so it does not need to disqualify a brand-new project from delegation.

  wrangler pages deploy --branch <name> remains on Pages because an interactive new-project flow separately prompts for its production branch. The deployment branch may therefore represent a preview and cannot safely be converted into a production Workers deployment.

• Widen agent Pages-to-Workers delegation to new projects on accounts that already use Pages

  When run by an AI agent, wrangler pages deploy and wrangler pages project create now delegate a brand-new static Pages project to a Workers static-assets deploy even when the account already has other Pages projects. The gate is now per-project rather than per-account: a command targeting a project that already exists stays on Pages, but a new project is delegated regardless of the account's other Pages projects.

  A project name restored from the Pages configuration cache is only used when the cache belongs to the currently authenticated account. An account-matching cached name remains on Pages even when the project is missing remotely, preserving the user's recorded Pages intent. After switching accounts, an otherwise unnamed deploy stays on Pages rather than treating a stale cached project name as a new project on the selected account.

Patch Changes

• Fix wrangler d1 execute --local being extremely slow with large SQL files or commands

  The local SQL splitter consumed quoted strings and comments character-by-character, re-checking the full accumulated string each time. This made splitting a large quoted value or comment quadratic, so seed files could take tens of seconds to run. The splitter now only inspects a bounded trailing window on each step, making splitting effectively linear. The remote path is unaffected as it imports the file server-side.

• Fix wrangler dev running the custom build command twice on startup and on every config change

  Wrangler already runs the custom build.command once before starting wrangler dev, to resolve the Worker's entry point. When dev.watch wasn't explicitly disabled, BundlerController then unconditionally ran the same build command again the moment it started watching for changes, and repeated this on every subsequent config reload too.

  For fast build commands this just meant duplicate log output (e.g. a vite build visibly running twice at startup). For slower or stateful build commands, running two builds concurrently against the same output files could corrupt the result or fail outright (for example, non-deterministic wasm-opt failures have been reported for Rust builds).

  The initial watcher setup now only bundles the output the build command already produced, instead of re-running the command. Real file changes detected by the watcher still re-run the build command as before.

Patch Changes

• Prevent Wrangler from exiting when a process capturing its output closes the pipe.

  Wrangler now ignores broken-pipe errors from stdout and stderr while preserving the existing failure behavior for other output errors.

• Update bundle size warning thresholds to use uncompressed size instead of gzip size

  The compressed script size limits (3 MiB free / 10 MiB paid) have been removed server-side in favor of a single 64 MiB uncompressed limit. The bundle size reporter now compares the uncompressed bundle size against this 64 MiB limit for its color-coded warnings, instead of comparing gzip size against the old 3 MiB compressed limit.

• Honor Workers Builds name overrides in wrangler preview

  Preview commands now target the Worker name supplied by Workers Builds instead of the name in local Wrangler configuration. This prevents preview builds from failing when the two names differ.

• wrangler dev no longer exits when a request to your Worker fails transiently

  Previously, a transient network failure on a single request — most commonly a request arriving just as an idle internal connection was closed, after roughly five seconds without traffic — could take down the whole dev server with an empty ✘ [ERROR], leaving the port unbound until restarted. In CI test suites, one such failure caused every remaining test to fail with connection errors.

  wrangler dev now automatically retries the affected request if it is safe to repeat (GET and HEAD requests). If a request still fails, it fails individually — the error is logged with the request method and URL — and the dev server keeps serving.

Minor Changes

• Support gzip compression for JSON Pipelines sinks

  Pipelines is in open beta. wrangler pipelines sinks create and the interactive setup flow now pass the selected JSON compression to the Pipelines API. JSON sinks accept uncompressed or gzip, while Parquet retains its existing compression options and zstd default.

• Add a --json flag to the wrangler workflows commands

  Every wrangler workflows command now accepts --json, which emits the raw API payload instead of the human-readable rendering. The formatted output remains the default, so existing usage is unaffected:

  wrangler workflows instances list my-workflow --json

  The JSON output carries raw values rather than a serialisation of the formatted view: ISO timestamps instead of locale-formatted dates, plain status strings instead of emojified labels, and no presentation-only derived fields.

Patch Changes

• Tolerate missing permissions during wrangler delete cleanup checks

  wrangler delete now warns and continues when it cannot inspect Worker dependencies or clean up legacy Workers Sites KV namespaces because of missing permissions. The Worker delete request itself still fails normally if the token cannot delete the Worker.

• Tolerate missing resource permissions during resource provisioning

  When Wrangler cannot check whether a bound resource exists because the API returns a 403, it now skips automatic provisioning for that resource type and continues the deploy. The deploy may still fail later if the resource is missing.

• Fix remote development with static assets for API tokens using granular Worker permissions

  Wrangler now creates Workers.dev preview sessions through the Worker-scoped endpoint and derives the preview hostname from the session response. This avoids requiring account-level Workers subdomain access.

Minor Changes

• Move binding utilities into @cloudflare/workers-utils

  Binding conversion, printing, and local-development validation are now exported from @cloudflare/workers-utils so they can be shared by Wrangler, the Cloudflare Vite plugin, and other consumers.

  The corresponding exports have been removed from @cloudflare/deploy-helpers. Consumers should import them directly from @cloudflare/workers-utils instead.

  Wrangler's unstable_printBindings API now accepts the bindings and an options object instead of five positional parameters.

• Add --date-start and --date-end filters to wrangler workflows instances list

  You can now narrow an instance listing to a creation-time window:

  wrangler workflows instances list my-workflow --date-start 2026-01-01 --date-end 2026-01-31

  Either flag can be used independently. Both accept an ISO 8601 date or timestamp and are normalised to UTC before being sent, so a date-only value such as 2026-01-01 works as well as a full 2026-01-01T13:00:00Z. The bounds are inclusive and compose with the existing --status filter.

• Add query string redaction to Workers observability configuration

  Set observability.redact_query_string in wrangler.json or observability.redactQueryString in the experimental cloudflare.config.ts format to remove query strings from request URLs in logs and traces.

• Include exact raw and gzip-compressed Worker bundle sizes in structured deploy and version-upload output.

Patch Changes

• Reduce the installed bundle sizes of Wrangler and Miniflare

  Wrangler now resolves bundled workspace dependencies from source during monorepo builds so unused exports can be removed. Miniflare, its shared CLI and container dependencies now use granular @cloudflare/workers-utils entry points instead of loading the package barrel, reducing the raw Wrangler and Miniflare artifacts by 6.16 MiB (31.4%) and 1.06 MiB (22.9%) respectively without changing runtime behavior or installed dependencies.

• Reduce Wrangler's published package size

  Stop including the unused build metafile in the npm package, reducing its unpacked size by approximately 3.1 MiB.

• Skip the skills install status lookup when telemetry is disabled

  Telemetry events include a currentAgentSkillsInstalled property, and computing it can query the GitHub API. The lookup used to start before the telemetry permission was checked, so users who opted out via WRANGLER_SEND_METRICS, DO_NOT_TRACK, or send_metrics in their Wrangler config still triggered network requests on behalf of telemetry. The dispatcher now checks the permission first and only performs the lookup when telemetry is enabled.

Minor Changes

• Add support for configuring a per-workflow max concurrency limit via workflows.concurrency.limit in your Wrangler config.

  The limit is the maximum number of Workflow instances that can run concurrently. It is validated as a positive integer and persisted on deploy; the ceiling is enforced server-side. Concurrency is ignored in local development.

  {
    "workflows": [
      {
        "binding": "MY_WORKFLOW",
        "name": "my-workflow",
        "class_name": "MyWorkflow",
        "concurrency": { "limit": 10 }
      }
    ]
  }

Patch Changes

• Upload wrangler preview modules as multipart form data

  wrangler preview used to base64 the bundle, its modules, and any sourcemaps into a single JSON request body. Base64 inflates content by a third, so a Worker with a large sourcemap could exceed the API request size limit and fail to deploy.

  The preview deployment request is now multipart/form-data. The deployment settings travel in a metadata part and each module follows as its own part carrying raw bytes, matching how wrangler deploy already uploads a Worker.

Minor Changes

• Add default_retention to Workflow bindings for configuring how long instances are retained

  Workflow instances are retained for an account-wide default period after they finish. You can now set a per-Workflow default in your Wrangler configuration, applied to instances that do not specify their own retention:

  {
    "workflows": [
      {
        "binding": "MY_WORKFLOW",
        "name": "my-workflow",
        "class_name": "MyWorkflow",
        "default_retention": {
          "success_retention": "3 days",
          "error_retention": "7 days"
        }
      }
    ]
  }

  Each side is optional and accepts either a duration string such as "3 days" or a whole number of milliseconds. Durations are interpreted by the Workflows API, which also caps them at your account's retention limit.

• Include a chronological list of handler events in email test harness results, so programmatic local email tests can assert the order in which messages are received, forwarded, replied to, or rejected.

  const result = await server.getWorker().email({
    from: "sender@example.com",
    to: "inbox@example.com",
    raw: [
      "From: Sender <sender@example.com>",
      "To: Inbox <inbox@example.com>",
      "Message-ID: <test@example.com>",
      "Subject: Test email",
      "",
      "Hello from the test harness",
    ].join("\r\n"),
  });
  
  expect(result.events).toEqual([
    { type: "received", timestamp: expect.any(String) },
    {
      type: "forward",
      timestamp: expect.any(String),
      messageId: expect.any(String),
    },
    {
      type: "reply",
      timestamp: expect.any(String),
      messageId: expect.any(String),
    },
  ]);

• Add experimental wrangler hyperdrive planetscale signature for provisioning Cloudflare-billed PlanetScale databases

  wrangler hyperdrive planetscale signature prints a signed authorization as JSON, proving to PlanetScale that Cloudflare will be billed for the database you are about to create:

  npx wrangler hyperdrive planetscale signature | \
    pscale database create <name> \
      --org <org> \
      --engine postgresql \
      --cloudflare-billing @- \
      --format json

  pscale database create defaults to Vitess, so pass --engine postgresql for a Postgres database, and --format json is recommended when the output is consumed by an agent.

  This requires pscale v0.313.0 or newer. Wrangler authorizes the Cloudflare billing side only, so your PlanetScale credentials stay between you and pscale.

  The signature is a cryptographically signed token that authorizes creating a database billed to your Cloudflare account. Treat it as a credential and do not share it. Piping it, as above, is recommended over passing it as a command line argument.

  This command is experimental and its interface may change.

• Enable FUSE-capable local container development

  Miniflare now automatically passes the Docker privileges needed for FUSE to local Durable Object containers when using local rootless Docker on Linux with /dev/fuse available, or a local Docker engine on macOS or through WSL where Linux containers run in a VM. This applies to Wrangler, the Cloudflare Vite plugin, and direct Miniflare use.

• Record the selected mode in the Build Output Specification top-level config.json

  The mode a build was produced in is now written to .cloudflare/output/v0/config.json as a mode field, alongside the account and compliance settings.

• Add pull request metadata to wrangler preview deployments

  wrangler preview now detects the pull request associated with the current CI run (GitHub Actions, GitLab CI, CircleCI, and a generic PULL_REQUEST_URL/PR_URL/CHANGE_URL fallback) and attaches it, along with the repository URL, to the preview deployment as annotations (workers/pull_request_number, workers/pull_request_url, workers/repository_url).

  This is best effort: if no pull request can be detected, nothing changes. When a pull request is detected, its URL is now also shown in the wrangler preview command output.

• Add pull request title to wrangler preview deployment annotations

  wrangler preview now also detects the title of the pull/merge request associated with the current CI run (GitHub Actions and GitLab CI, plus a generic PULL_REQUEST_TITLE fallback) and attaches it to the preview deployment as the workers/pull_request_title annotation, alongside the existing pull request number/URL, repository URL, and commit SHA annotations.

  This is best effort: if no pull request title can be detected, nothing changes.

Patch Changes

• Prepare autoconfig for multiple configuration targets

  Add target-specific configuration output and command detection while preserving Wrangler's existing setup and deployment behavior.

• Fix wrangler login --use-keyring incorrectly reporting that secret-tool is missing on Linux

  Libsecret's secret-tool does not support --version; it prints usage and exits 2, which Wrangler previously interpreted as unavailable. Wrangler now reports it missing only when launching the executable fails.

• [private beta]: Explain unavailable Preview URLs after wrangler preview deployments

  When a Preview deployment has no active URLs, Wrangler now explains how to enable Preview Deployments on workers.dev or a custom domain.

• Stop automatically offering to install Cloudflare skills for new users

  Wrangler will no longer prompt new users to install Cloudflare skills after commands complete. It will continue to offer updates to skills that Wrangler previously installed.

Minor Changes

• Add connect trigger for raw sockets

  You can now configure a Worker to receive raw socket connections during wrangler dev, delivered directly to the Worker's connect(socket, env, ctx) handler:

  {
    "connect": [{ "protocol": "tcp", "port": 5432 }]
  }

  Each entry opens a listening socket on 127.0.0.1 (or the given address) that forwards incoming connections straight to the Worker, bypassing the local dev HTTP entry point. This requires the experimental compatibility flag. Only "tcp" is supported at the moment.

  @cloudflare/config also supports declaring this trigger via triggers.connect(...), which lowers to the connect field above:

  import { defineWorker, triggers } from "@cloudflare/config";
  
  export default defineWorker({
    triggers: [
      triggers.connect({ protocol: "tcp", port: 5432, address: "127.0.0.1" }),
    ],
  });

• Add container support to worker previews

  Worker previews now support containers through a new previews.containers configuration block. Container configuration doesn't inherit, so declare containers explicitly in the previews block to enable them for previews. This mirrors how previews.durable_objects works today. Wrangler names each preview container application {worker_name}_{preview_slug}_{class_name}, normalising and shortening the result to what the API accepts. Either change appends a short digest of the composed name, so two names that would otherwise land on one stay distinct. An entry cannot set its own name, because application names are unique to an account and a fixed name would collide between two previews of the same Worker. A Durable Object class is backed by at most one container application, so the validator rejects two entries that share a class_name. Wrangler skips container applications bound to Durable Object classes that another Worker implements through script_name, because the implementing Worker owns its own container application. A binding is not required: a Durable Object declared through migrations or exports and reached only over ctx.exports can still back a container. Every entry must set class_name. A previews.containers entry whose class_name matches no Durable Object class at all is rejected before the preview deployment is created, so a typo fails loudly instead of producing a preview with no container.

  Wrangler creates the container applications on wrangler preview. Deleting a preview tears them down server side, so wrangler preview delete doesn't remove them.

  Container build and deploy progress prints to stdout. wrangler preview --json suppresses wrangler's own output so it doesn't interleave with the payload, and warnings and errors still go to stderr. Docker's build output and the progress spinner write to stdout directly and bypass that suppression, so parse --json from a non interactive shell, where the spinner is skipped, and prefer a prebuilt image over a Dockerfile.

• [private beta]: Create the parent Worker automatically when wrangler preview targets one that doesn't exist yet

  Previews hang off a parent Worker, so running wrangler preview before the Worker had ever been deployed failed with a raw API error naming the Preview endpoint. Wrangler now offers to create an empty parent Worker and then carries on creating the Preview. The parent uses the same workers.dev and Preview URL settings that wrangler deploy would resolve, without applying routes or cron triggers. In non-interactive environments, Wrangler creates the Worker without asking.

• Add individual and batch Workflow instance deletion to the runtime and SDK.

  • WorkflowInstance.delete() deletes one instance. Self-deletion stops the current execution.

  • env.MY_WORKFLOW.deleteBatch(instanceIds) deletes up to 100 instances and returns { deleted, errors } per input position.

  • wrangler workflows instances delete <name> [id..] deletes instances remotely or with --local; IDs can also come from a JSON array passed with --filename, with a combined limit of 100.

Patch Changes

• Fixes kv bulk put corrupting binary values written to local KV

  Values marked base64: true were stored incorrectly whenever they contained bytes that do not form valid UTF-8, which covers images, compressed data and most other binary payloads. A Worker reading such a key back under wrangler dev got a different, longer value than the one that was written: a 12 byte PNG header came back as 20 bytes.

  kv bulk put writes to local KV by default, so the plain command was the affected one. Remote writes were never affected, and neither were entries without base64 or values written with kv key put.

• Move deploy output writing into shared deploy helpers

• Remove unsupported remote configuration from Workflow bindings

  Workflow bindings no longer accept remote in configuration, as remote Workflow bindings have never actually been supported.

• Fix the --temporary error on commands that authenticate more than one time

  wrangler d1 migrations apply --remote --temporary failed with this error: You're already authenticated with Cloudflare, so --temporary can't be used. The failure occurred with no login and with no CLOUDFLARE_API_TOKEN. This command authenticates one time for each statement that it runs. The first authentication makes a temporary preview account. The second authentication read the token of this new account as an earlier login.

  Wrangler now uses again the temporary account from the same command run. Commands that authenticate more than one time now work as wrangler deploy --temporary works. If real credentials are available, --temporary is still an error.

Minor Changes

• Allow containers to be attached to a Durable Object from its exports entry

  A container can now be linked to its Durable Object from the export side, using a new container field that names an entry in the containers array. As a result containers.class_name is now optional — a container that is referenced this way only needs a name:

  {
    "name": "my-worker",
    "main": "worker.js",
    "compatibility_date": "2026-07-01",
    "containers": [
      { "name": "my-container", "image": "./Dockerfile", "max_instances": 1 }
    ],
    "exports": {
      "MyContainerDO": {
        "type": "durable-object",
        "storage": "sqlite",
        "container": "my-container"
      }
    }
  }

  The existing containers.class_name direction keeps working and either direction may be used, but the two must agree: a container that names its Durable Object cannot also be claimed by a different one.

  container is only valid on live durable-object exports (created and expecting-transfer) and requires storage: "sqlite". Wrangler now also reports an error when:

  • a container reference names a container that does not exist

  • two Durable Object exports claim the same container

  • a container and a Durable Object export disagree about which one they are linked to

  • a container ends up linked to no Durable Object at all

  • two containers share a name

  • a container's class_name names a Durable Object whose storage is legacy-kv

  • two containers are attached to the same Durable Object

  That last case was previously accepted but could never work: workerd attaches a single container per Durable Object namespace, and in local development every container for a class builds into the same image tag, so one silently overwrote the other. If you have two containers on one class_name, give each its own Durable Object class.

Patch Changes

• Honor access.dev when running Workers with @cloudflare/vitest-pool-workers, so ctx.access.getIdentity() returns the configured identity just as it does with wrangler dev.

• Fix .env loading on Windows leaking stale, differently-cased duplicate keys

  On Windows, wrangler loads .env values through a case-insensitive Proxy wrapper so lookups like env.PATH and env.Path resolve to the same value, and this object is assigned directly to process.env. When a key was set again under a different casing (e.g. a value in .env.local overriding one from .env with different casing), the previous casing was never removed from the underlying object. env.PATH/env.Path still returned the correct, latest value, but anything that enumerates process.env — Object.keys, for...in, JSON.stringify, object spread, or a spawned subprocess inheriting the environment — would see both the stale and current key.

  Duplicate entries no longer appear, so environment variables passed to subprocesses and any code that lists the environment now see only the latest value for each variable.

• Normalize structural CRLF line endings before sending D1 commands to the remote query API

  wrangler d1 migrations apply --remote and wrangler d1 execute --remote --command failed with incomplete input: SQLITE_ERROR when the SQL contained CRLF line endings inside a compound statement such as a CREATE TRIGGER ... BEGIN ... END; body. Structural line endings are now normalized to LF before the command is sent to the D1 query API, while CRLF inside quoted values and identifiers remains unchanged.

• Fixes D1 SQL statements not handling lowercase ends correctly

  wrangler d1 execute and wrangler d1 migrations apply split a SQL file into statements before running them. A BEGIN or CASE block closed with a lowercase end was not recognised as closed, so every statement after it was folded into that block instead of being run on its own. SQLite accepts either case, so a file like this applied only the trigger and silently skipped the table:

  CREATE TRIGGER IF NOT EXISTS update_trigger AFTER UPDATE ON items
  begin
      DELETE FROM updates WHERE item_id=old.id;
  end;
  CREATE TABLE after_the_trigger (id TEXT PRIMARY KEY);

  Files written with an uppercase END were unaffected. Both cases now behave the same.

• Resolve --latest to the newest compatibility date supported by the installed runtime

  wrangler deploy --latest and wrangler versions upload --latest resolved the compatibility date to the current date, and wrangler pages download config did the same for projects configured to always use the latest compatibility date. Both write that date into a configuration file for subsequent commands to use, so a date that the installed workerd did not yet support left the project unable to run wrangler dev.

  These now resolve to the latest compatibility date supported by this version of Wrangler, which is the release date of the workerd it ships with.

• Fix spurious Trailing comma jsonc(519) warnings for wrangler.jsonc in VS Code 1.131+

  Trailing commas in wrangler.jsonc files that reference Wrangler's JSON schema are no longer reported as errors by recent versions of VS Code. Wrangler always accepted these files; only the editor warning was wrong.

• Respect CLOUDFLARE_ACCOUNT_ID in wrangler pages project list, create and delete

  These three commands could target a previously used account even when CLOUDFLARE_ACCOUNT_ID was set, failing with Authentication error [code: 10000] in setups with more than one account. They now use the account named by CLOUDFLARE_ACCOUNT_ID, matching the rest of wrangler pages. When the variable is unset, the previously used account is still selected, as before.

• Fix wrangler preview base-config commands showing an inherited script positional

• Use a fixed default compatibility date rather than the current date

  When no compatibility date was set, Wrangler, C3 and the Vitest pool all defaulted to the current date. workerd only accepts a compatibility date up to 7 days beyond its own release, so whenever a workerd release was delayed the default could get ahead of the runtime that had been installed, and local development would fail to start.

  The default is now fixed at the release date of the workerd version that ships with each release, which leaves a week of headroom and updates as workerd is upgraded. @cloudflare/vite-plugin previously inlined the date at which it was built. It now shares the same default.

• Prevent date-enabled Node.js compatibility from adding conflicting globals to generated runtime types

  Runtime type generation now treats Node.js compatibility enabled by a compatibility date the same way as an explicit nodejs_compat flag. Node.js globals continue to come from @types/node instead of being generated as any declarations that override those types.

• Use the FedRAMP High managed container registry when Wrangler targets the FedRAMP High compliance region

  Container builds, pushes, deployments, image commands, and local development now select the corresponding production or staging FedRAMP registry and API from either compliance_region or CLOUDFLARE_COMPLIANCE_REGION.

• Enable the new configuration format in the cf-wrangler dev delegate

  Projects started through cf dev now load cloudflare.config.ts and optional wrangler.config.ts, matching the configuration used by the delegate's build path.

Minor Changes

• Add local dev simulation for Cloudflare Access ctx.access.getIdentity()

  You can now configure a mock Cloudflare Access identity in wrangler.json so that ctx.access.getIdentity() returns it during local development.

  // wrangler.json
  {
    "access": {
      "dev": {
        "aud": "my-app-aud-tag",
        "identity": {
          "email": "user@example.com",
          "name": "Test User"
        }
      }
    }
  }

• [private beta]: Updates the --ignore-defaults flag to --ignore-base-config on wrangler preview commands.

  --ignore-base-config now only takes effect on Preview creation, rather than on each deployment, since Preview base configuration is now copy-on-create rather than inherit-on-deploy.

• Add automatic update prompts for out-of-date Cloudflare agent skills

  When Cloudflare skills were previously installed by Wrangler and the upstream cloudflare/skills repository has newer content, Wrangler now offers to update them after eligible commands complete.

  To reduce prompt fatigue, the update check only runs once a month (30 days since the last install or update). Declining suppresses the prompt until the next upstream change.

  When declining an update, Wrangler offers the option to permanently disable future update prompts. This preference is stored globally in ~/.wrangler/agents-skills-install.jsonc. The WRANGLER_NO_SKILLS_UPDATE_PROMPTS=true environment variable can also be used to suppress prompts. The --install-skills flag remains available regardless of these settings.

Minor Changes

• Detect Node.js compatibility from the compatibility date, now that nodejs_compat is enabled by default

  As of compatibility date 2026-08-04, workerd enables the nodejs_compat and nodejs_compat_v2 compatibility flags by default. Previously these tools only treated Node.js compatibility as enabled when one of those flags was listed explicitly, so a Worker on a compatibility date of 2026-08-04 or later without the flag would get Node.js APIs from the runtime but no Node.js polyfills from the bundler, and process.env could be substituted with an empty object at build time. They now resolve these flags the same way workerd does, and honour no_nodejs_compat to opt out.

  To keep Node.js compatibility switched off on a newer compatibility date, specify both no_nodejs_compat and no_nodejs_compat_v2, since each flag has its own default.

  @cloudflare/vitest-pool-workers needs nodejs_compat_v2 for its own test runner, so it continues to override a project that opts out of it. On a compatibility date that enables the flag anyway, it now drops the opt-out rather than adding the flag back, which workerd would reject — previously this stopped such a project from running any tests at all.

  wrangler types also no longer attributes its @types/node suggestion to "the nodejs_compat flag", which it can now make for Workers that do not set the flag at all.

Patch Changes

• Ignore a nodejs_compat compatibility flag that the compatibility date already enables

  workerd rejects a compatibility flag that its compatibility date enables by default, so a Worker configured with both a compatibility date of 2026-08-04 or later and nodejs_compat failed to start locally with "The compatibility flag nodejs_compat became the default as of 2026-08-04 so does not need to be specified anymore".

  The redundant nodejs_compat and nodejs_compat_v2 flags are now dropped when starting the runtime, which has no effect on the resulting Worker because the compatibility date enables both anyway. no_nodejs_compat and no_nodejs_compat_v2 still switch Node.js compatibility off, and a flag specified alongside its own opt-out is left alone so that workerd still reports those as contradictory.

• Stop adding a redundant nodejs_compat flag to generated Wrangler configurations

  create-cloudflare and wrangler setup write today's date as the compatibility_date, and from 2026-08-04 that already enables nodejs_compat. Adding the flag as well made the generated project fail to start with "The compatibility flag nodejs_compat became the default as of 2026-08-04 so does not need to be specified anymore", so the flag is now only added for earlier compatibility dates.

  create-cloudflare also removes the flag when a template, or a framework's own scaffolder, already wrote it into a configuration that ends up using such a compatibility date, and still installs @types/node for these projects even though there is no longer a flag to detect them by.

  wrangler setup does the same for a wrangler.json(c) that is already in the project: it writes today's date over whatever date that configuration was written for, so a nodejs_compat it finds there is removed as part of writing the file.

• Fix remote binding sessions reusing stale binding configurations

  Starting a new remote bindings session that reuses a Worker name no longer picks up the bindings from a previous session, which could cause Binding "..." not found errors.

Minor Changes

• Add Preview base config secret commands

  Wrangler now manages Worker Preview base config secrets with wrangler preview base-config secret put, delete, list, and bulk. These commands update the Worker's previews_base_config.env, keeping shared defaults scoped to all of that Worker's Previews. wrangler preview base-config secret list reads from the Worker's Preview base config and prints secret names with values masked. wrangler preview base-config secret bulk deletes a secret when its value is null, matching wrangler secret bulk.

• Allow Wrangler projects to build a Worker once and reuse it in createTestHarness()

  Build the Worker once:

  wrangler deploy --dry-run --outdir ./worker-output

  Then reuse the emitted Worker during test harness startup and reset:

  const server = createTestHarness({
    workers: [
      {
        configPath: "./wrangler.jsonc",
        prebuiltWorkerDir: "./worker-output",
      },
    ],
  });

• Add email.sending as an event subscription source for queues

  wrangler queues subscription create now accepts --source email.sending alongside two new flags, --zone-id and --domain, which identify the zone and the sending domain (zone apex or a verified subdomain) to subscribe to. Both flags are required for this source. The subscription's resource is displayed as the sending domain in wrangler queues subscription get.

• Add US jurisdiction support to wrangler d1 create

  You can now create a D1 database in the US jurisdiction with wrangler d1 create <name> --jurisdiction us. The new jurisdiction is also listed in the command's help output.

• Use Preview deployment PATCH APIs for Preview secret commands

  Wrangler now updates Worker Preview secrets by patching the named Preview's latest deployment instead of patching the Worker's Previews settings. This keeps secret changes scoped to one Preview, avoids affecting production or other Previews, and creates a new Preview deployment that goes live at 100% immediately. wrangler preview secret list now reads from the named Preview's latest deployment and prints secret names with values masked. wrangler preview secret bulk now deletes a secret when its value is null, matching wrangler secret bulk.

• Honor DO_NOT_TRACK=1 as a telemetry opt-out

  Wrangler now disables telemetry when DO_NOT_TRACK=1 is set, regardless of other telemetry settings.

Patch Changes

• Compact wrangler preview deployment success output

  wrangler preview now prints a concise success summary with the Preview name, Preview URL, deployment ID, and Deployment URL instead of the previous box-art settings summary.

• Fetch script metadata directly instead of listing all scripts

  When resolving Durable Object migrations, fetch the specific script's service metadata via /workers/services/{name} instead of listing all scripts in the account via /workers/scripts. This avoids downloading metadata for every Worker in the account just to find one script's migration tag.

• Fix wrangler dev commands crashing with No such module "wrangler:modules-watch" when "no_bundle": true

  Running wrangler dev or wrangler pages dev with bundling disabled ("no_bundle": true in wrangler.json, or the --no-bundle flag) no longer crashes at startup with Uncaught Error: No such module "wrangler:modules-watch". Live reloading on file changes continues to work as before.

Patch Changes

• Update local development for Miniflare's config-based options

  Wrangler now converts the Miniflare options it creates for local development to Miniflare's config-based workers shape.

  Users should not expect to notice any changes.

Minor Changes

• Adds the ability to find container instances by exact ID or name

  wrangler containers instances <application_id> --search <instance_id_or_name> now searches every page and returns exact matches in human-readable or JSON output. JSON returns a top-level array, including an empty array when there is no match, while human-readable output prints a no-match message. If multiple instances have the same exact name, every matching instance is returned.

• Add explicit pagination to container instance JSON output

  Use wrangler containers instances <application_id> --json --per-page <size> to return one page with machine-readable result_info, then pass its next_page_token to --page-token to retrieve the next page. Plain --json remains backward-compatible: it requests the complete list and returns the existing top-level array.

Patch Changes

• Cut the per-request cost of local observability capture

  Every tail event was written to the trace store as its own Durable Object call, so a request paid two or three round-trips per span. On a module-heavy app under the Vite plugin that dominated dev request latency. Rows are now buffered and written in batches, taking a request from roughly thirty calls to three.

  Work in progress still shows up as it happens: the root span is written immediately, console logs and exceptions as they arrive, and a span's completion is written on the next event once 100ms has passed. An invocation that goes completely quiet writes nothing further until it ends, since the flush is driven by tail events rather than a timer.

  The Vite plugin's own router, asset and proxy workers are also no longer captured. Their traces were noise the Observability views already hid, and skipping them cuts the spans recorded per request — a side benefit being that a trace's root is now your Worker rather than __router-worker__.

Minor Changes

• Add --parse-type flag to wrangler ai-search create

  wrangler ai-search create now accepts --parse-type to control how a website data source discovers URLs. sitemap (the default) reads XML sitemaps; discover follows links recursively.

  Previously the parse type could only be chosen through the interactive wizard, which was skipped whenever --source was supplied — so it was impossible to create a discover instance from a script.

  wrangler ai-search create my-instance \
    --type web-crawler \
    --source https://example.com \
    --parse-type discover

  The interactive wizard now offers Discover alongside Sitemap. --parse-type is only valid with --type web-crawler; passing it with --type builtin or --type r2 is rejected, since the API stores the value for those source types but never reads it. When the flag is omitted in non-interactive mode the field is left unset and the API default (sitemap) applies.

• Improve the Local Explorer's Observability views

  console.log messages now render the way the console would (JSON-encoded strings are unwrapped and multi-argument logs are joined), traces and events can be looked up by trace or span id from the search bar, and an event's "View trace" button jumps to the exact invocation that emitted it — even when a trace_id spans several invocations (e.g. a subrequest or self fetch).

• Add support for OAuth 2.0 Device Authorization Grant to wrangler login

  Run wrangler login --device to authenticate without a local callback server. Useful in containers, remote SSH sessions, Codespaces, and any other environment where localhost:8976 is unreachable from your browser.

  The new flow:

  • prints the verification URL and user code to the terminal,

  • attempts to open the verification URL in your default browser automatically (suppressed via --browser=false),

  • and polls the token endpoint until you approve the request (with a 5-minute hard cap).

  The verification URL is supplied by the authorization server, so it is rejected unless it is an https URL on the same auth domain the device code was requested from — it is never printed or opened otherwise.

  --callback-host and --callback-port cannot be combined with --device, since this flow does not start a local callback server.

Minor Changes

• Add --json flag to wrangler containers info for consistent JSON output with sibling commands list and instances

• Enable local observability capture by default in dev

  wrangler dev and the Vite plugin now capture request traces and console logs into the Local Explorer's Observability tab out of the box — previously this was opt-in behind X_LOCAL_OBSERVABILITY=true. Set X_LOCAL_OBSERVABILITY=false to opt out (for example if the extra per-worker collector/streaming-tail services cause trouble in a multi-process dev-registry setup).

• Add additional triggers to Workflows

  Workers can now declaratively start a locally defined Workflow. Configure event subscriptions under triggers.events; Wrangler validates each target and updates the script's event triggers during deployment.

  {
    "triggers": {
      "events": [
        {
          "type": "cf.artifacts.repo.pushed",
          "filter": {
            "namespace": "my-namespace",
            "repo_name": "my-repo"
          },
          "targets": [
            {
              "type": "workflow",
              "workflow_name": "my-workflow"
            }
          ]
        }
      ]
    }
  }

Patch Changes

• Fix jsx_fragment being ignored when wrangler dev runs a custom build

  If your project uses a custom build and sets both jsx_factory and jsx_fragment, wrangler dev used your jsx_factory value for JSX fragments as well, so fragments compiled incorrectly. Your jsx_fragment value is now used.

• Stop wrangler dev starting new work after you stop it or it reloads

  Stopping wrangler dev, or having it reload after a configuration change, could still leave it starting work for the state it had just left behind: your custom build command could run once more after dev had stopped, a change to a file in your assets directory could be reported against configuration that had already been replaced, and in some cases the process could stay alive instead of exiting.

  That work is now discarded, so stopping or reloading wrangler dev leaves nothing running behind it.

• Stop wrangler dev from running custom builds concurrently

  When several watched files changed at once — for example during a git pull or a "save all" — wrangler dev started a custom build for every file that changed, so multiple copies of your build command ran at the same time and fought over the same output files.

  A burst of file changes now results in a single build, and a build only starts once the previous one has finished.

• Report a clear error for account IDs that can't be used in a Cloudflare API request

  Account IDs are substituted straight into Cloudflare API URL paths, so a value containing non-ASCII characters previously failed deep inside the request layer with an opaque Cannot convert argument to a ByteString error that gave no hint about which setting was at fault. Account IDs read from CLOUDFLARE_ACCOUNT_ID and from the account_id configuration field are now validated up front, and an invalid value fails with a message naming both the offending value and where it came from.

Minor Changes

• removed

  Remove containerEngine from the worker options returned by unstable_getMiniflareWorkerOptions

  unstable_getMiniflareWorkerOptions no longer includes containerEngine in the returned workerOptions, since the container engine is a Miniflare instance-wide setting rather than a per-worker one. Callers that build a Miniflare instance from these options should set containerEngine at the top level instead.

Patch Changes

• Rewrite local testing paths (/cdn-cgi/*)

  Miniflare v5 moved its internal local testing endpoints to /cdn-cgi/local/* (and /__cf_local/* for endpoints that must remain reachable over tunnels) to prevent any potential collision with production routes. wrangler dev and the Vite plugin now transparently rewrite the old paths to the new ones, meaning you can continue to use the old paths without issue.

  These are the new paths:

  • /cdn-cgi/handler/scheduled → /cdn-cgi/local/scheduled

  • /cdn-cgi/handler/email → /cdn-cgi/local/email

  • /cdn-cgi/explorer/* → /cdn-cgi/local/explorer/*

  • /cdn-cgi/mf/scheduled → /cdn-cgi/local/scheduled (Note /cdn-cgi/mf/scheduled is already deprecated)

  • /cdn-cgi/mf/stream/* → /__cf_local/stream/*

  • /cdn-cgi/mf/imagedelivery/* → /__cf_local/imagedelivery/*

Minor Changes

• Avoid Worker and workers.dev naming prompts in agent-driven deploys

  Wrangler now derives the Worker name from the project and automatically registers the same project-derived workers.dev account subdomain on a first deploy when running in a detected agent environment. The deploy output explains how to change both names.

• The experimental build output directory now includes the Worker's configuration at .cloudflare/output/v0/workers/default/config.json instead of .cloudflare/output/v0/workers/<worker-name>/worker.config.json

• Graduate wrangler check startup from alpha and show bundle size and a local timing summary

  The command no longer prints an alpha warning. It now reports its local profile window, sampled active, garbage collection, and idle time alongside the raw and compressed bundle sizes. The existing measurement warning continues to distinguish these local measurements from startup time measured on Cloudflare.

• Add support for dispatching email handlers with createTestHarness

  You can now call server.getWorker().email({ from, to, raw }) to dispatch directly to a Worker's email() handler and inspect its outcome, rejection reason, forwarded messages, and replies.

  const result = await server.getWorker().email({
    from: "sender@example.com",
    to: "inbox@example.com",
    raw: [
      "From: Sender <sender@example.com>",
      "To: Inbox <inbox@example.com>",
      "Message-ID: <test@example.com>",
      "Subject: Test email",
      "",
      "Hello from the test harness",
    ].join("\r\n"),
  });
  
  expect(result).toMatchObject({
    outcome: "ok",
    forwards: [{ recipient: "archive@example.com" }],
    replies: [
      {
        sender: "inbox@example.com",
        raw: expect.stringContaining("Thanks for your email"),
      },
    ],
  });

Patch Changes

• Fix ctrl+c not being able to interrupt wrangler while waiting for Cloudflare Access authorization

  When a domain is behind Cloudflare Access (for example during remote bindings startup), wrangler runs cloudflared access login, which only returns once the user completes the authorization flow in the browser. This was invoked synchronously, blocking Node's event loop, so wrangler could not react to ctrl+c (or anything else) until the authorization completed — abandoning the browser flow left a hung wrangler process that had to be killed externally. cloudflared is now spawned asynchronously, keeping wrangler responsive while it waits. The remote runtime passes its abort signal through to the spawn, so tearing down the session kills a still-pending cloudflared immediately, with process exit as a last-resort cleanup.

• Include the local observability query endpoint in the agent-facing Local Explorer hint

  The hint wrangler dev prints for AI-agent sessions now lists POST /cdn-cgi/explorer/api/local/observability/query, so agents can discover the read-only SQL endpoint for captured request traces and console logs (the spans and logs tables) alongside the existing binding and storage routes.

• Improve the agent-facing Local Explorer hint for the observability query endpoint

  When a wrangler dev session is detected as running inside an AI agent, the hint for POST /local/observability/query now explains that the endpoint takes a read-only SQL query (SELECT/WITH only) over the captured spans and logs tables, notes that attributes is JSON (read via json(attributes)), and includes a copy-pasteable curl example. The full OpenAPI schema is demoted to a last-resort footer so agents reach for the small, actionable example first instead of fetching the large schema.

• Fix wrangler triggers deploy to use Vite-generated redirected configuration

  The command now reads .wrangler/deploy/config.json, matching wrangler deploy and wrangler versions upload, so generated Worker names and trigger settings are applied.

Minor Changes

• Add hidden --jurisdiction option to wrangler kv namespace create for internal testing

  This option creates a KV namespace within a specific jurisdiction (for example us, eu, or fedramp), backing it with jurisdiction-scoped storage. It is experimental and currently gated to allow-listed accounts, so it is hidden from --help until the feature is generally available.

• Add experimental local_dev.experimental_s3_credentials to r2_buckets config

  When set, the R2 bucket is served over a local S3-compatible API at /cdn-cgi/local/r2/s3/<bucket-id> during local development, authenticated with the configured AWS SigV4 credentials. <bucket-id> is the bucket's bucket_name, or the binding name if bucket_name is not set:

  {
    "r2_buckets": [
      {
        "binding": "BUCKET",
        "bucket_name": "my-bucket",
        "local_dev": {
          "experimental_s3_credentials": {
            "accessKeyId": "local-access-key-id",
            "secretAccessKey": "local-secret-access-key"
          }
        }
      }
    ]
  }

• Respect and surface the Retry-After header on Cloudflare API responses

  Previously, if a Wrangler command (e.g. wrangler versions upload, wrangler deploy) hit the Cloudflare API's rate limit, the resulting error gave no indication of how long to wait before trying again, and 429 responses weren't retried at all (only 5xx errors were, with a fixed linear backoff).

  Now:

  • 429 Too Many Requests responses are automatically retried, alongside the existing 5xx retry behaviour.

  • If a retried response includes a Retry-After header, Wrangler waits for that duration instead of the default backoff, and logs a message indicating how long it's waiting. To avoid blocking for an excessive amount of time, waits longer than 60 seconds fail fast instead — the surfaced Retry-After value lets the caller schedule its own retry.

  • If a retryable error is ultimately surfaced to the user (e.g. because retries were exhausted), the error message includes a note with the Retry-After duration, and the command-failed entry written to the Wrangler output file (WRANGLER_OUTPUT_FILE_PATH/WRANGLER_OUTPUT_FILE_DIRECTORY) gains a retry_after_ms field. This lets scripts and CI/CD pipelines calling Wrangler repeatedly (for example, wrangler versions upload on every commit) read the wait duration directly instead of regex-parsing stderr.

  APIError.isRetryable() is unchanged (still 5xx only); retryOnAPIFailure() separately retries 429s. retryAfterMs, when present, is honoured for any retried error, not just 429s.

  retryAfterMs is also now populated on APIErrors raised from direct R2 object requests, the Browser Rendering API, and errors surfaced from commands using the official cloudflare SDK client.

• Support connect() on remote VPC Network and VPC Service bindings in local development

  Remote VPC Network and VPC Service bindings previously only supported HTTP and JSRPC, so calling binding.connect(address) against a private TCP service (for example a database) failed in local dev with Incoming CONNECT on a worker not supported. Raw TCP connections through remote VPC Network and VPC Service bindings now work in local development.

  This feature is experimental. Existing HTTP and JSRPC usage of remote VPC Network and VPC Service bindings is unaffected, and no new configuration is required.

Patch Changes

• Color Email Routing plan change markers

  Wrangler now highlights additions in green, updates in yellow, and deletions and conflicts in red so Email Routing deployment plans are easier to scan before confirmation.

• Apply Email Routing changes across independent zones concurrently

  Wrangler now limits concurrent zone updates while preserving the Email Routing plan order within each zone. Deployments that configure addresses across multiple zones complete faster without breaking delete-before-add transitions at a zone's rule limit.

• Fix wrangler cloudchamber curl mangling header values that contain a colon

  Header values were split on every colon and only the segment between the first and second was sent, so --header location:https://example.com/x arrived as https. Headers are now split on the first colon only. A header that is not in the documented --header <name>:<value> form previously threw an unhandled TypeError, and now reports a clear error.

• Handle and explain authentication failures from remote bindings during local development

  Wrangler now recognizes authentication failures from remote preview sessions and reports that bindings which need to run remotely require Cloudflare authentication even when the rest of the Worker is developed locally.

• Speed up old debug log cleanup by reading each log file's date from its filename instead of stat-ing every file

  Wrangler periodically deletes debug log files older than 30 days from its logs directory. Previously it made a filesystem stat call for each file to read its modification time; it now derives the age from the timestamp already encoded in the log filename, avoiding that extra work.

Minor Changes

• Add local-dev observability

  wrangler dev and the Vite plugin now capture a trace for every local Worker invocation - spans, logs, and console.* output, including requests that cross worker or Durable Object boundaries.

  You can explore this data two ways:

  • A new Observability tab in the Local Explorer, with a Traces view (recent invocations, an inline timeline waterfall, and filters) and an Events view.

  • A read-only SQL endpoint at /cdn-cgi/explorer/api/local/observability/query, discoverable via the Local Explorer's OpenAPI document, so coding agents and tools can query the same spans and logs tables.

  While this is in testing it's off by default; set X_LOCAL_OBSERVABILITY=true to turn it on. It will be on by default in the public release.

Patch Changes

• Improve the D1 database-limit error message

  When creating a D1 database fails because the account has hit its database limit, the error now points to the relevant next steps — upgrading on the Workers Free plan or requesting a higher limit on a paid plan — alongside the existing commands to list and delete databases. Previously it only suggested deleting unused databases. This applies both to wrangler d1 create and to the D1 database that is created during resource provisioning on deploy.

• Fix grammar in the container image-too-large error

  The error thrown when a container image exceeds the available disk size ended with "Your need more disk for this image." It now reads "You need more disk for this image."

• Ignore the removed legacy_env field when reading a redirected configuration

  Older versions of tools such as the Vite plugin can generate a redirected configuration (.wrangler/deploy/config.json) that still includes the removed legacy_env field. Since these files are tool-generated, users could not easily remove the field themselves, and Wrangler would error out. Wrangler now silently strips legacy_env from redirected configurations. User-authored configurations still report an error so that the field can be removed.

• Fix dev proxy silently hanging or returning a misleading 503 on network errors for non-root-path requests

  During wrangler dev, a transient network error on any request path other than / could be misclassified as the worker being reloaded, even when it wasn't: GET/HEAD requests would silently hang (with nothing logged) until the client timed out, and other methods would receive a misleading Your worker restarted mid-request 503. Such errors are now reported and surfaced immediately when the worker has not actually changed.

• Explain how to provision Flagship bindings if app_id missing in remote development

  Wrangler now reports that a Flagship binding without an app_id must first be created with wrangler flagship apps create.