chore(deps): bump undici and @cloudflare/vitest-pool-workers

[dependabot]

Bumps undici to 7.28.0 and updates ancestor dependency @cloudflare/vitest-pool-workers. These dependencies need to be updated together.

Updates undici from 7.18.2 to 7.28.0

Release notes

Sourced from undici's releases.

v7.28.0

⚠️ Security Release

This release line addresses 7 security advisories, all shipped in v7.28.0.

Action required: Upgrade to undici 7.28.0 or later.

npm install undici@^7.28.0

The v7 line is not affected by GHSA-38rv-x7px-6hhq (CVE-2026-9675), which is an 8.x-only regression.

Note on GHSA-hm92-r4w5-c3mj: this fix shipped in v7.28.0, not the earlier 7.2x line — the vulnerable single-pool code was still present through v7.27.2. The per-origin pool fix is 3805b8f8 (#5041).

Summary

Advisory	CVE	Severity (CVSS)	Fixed in	Fix commit
GHSA-vxpw-j846-p89q	CVE-2026-12151	High (7.5)	7.28.0	8cb10f98
GHSA-vmh5-mc38-953g	CVE-2026-9697	High (7.4)	7.28.0	04201f89
GHSA-hm92-r4w5-c3mj	CVE-2026-6734	High (7.5)	7.28.0	3805b8f8
GHSA-pr7r-676h-xcf6	CVE-2026-9678	Moderate (5.9)	7.28.0	85a24055
GHSA-p88m-4jfj-68fv	CVE-2026-9679	Moderate (5.9)	7.28.0	d0574cc4
GHSA-g8m3-5g58-fq7m	CVE-2026-11525	Low (3.7)	7.28.0	d0574cc4
GHSA-35p6-xmwp-9g52	CVE-2026-6733	Low (3.7)	7.28.0	ea8930cf

---

High severity

WebSocket DoS via fragment count bypass — CVE-2026-12151

GHSA-vxpw-j846-p89q · CWE-400, CWE-770 Fix: 8cb10f98 websocket: limit the number of fragments in a message (part of backport a027a4a0 Backport WebSocket maxPayloadSize fixes to v7.x, #5423)

A malicious WebSocket server can stream a large number of small or empty continuation frames. Undici enforced a limit on cumulative payload size but did not limit the number of fragments per message, leading to unbounded memory growth and denial of service.

• Affected: applications using new WebSocket(...) or WebSocketStream against untrusted endpoints.
• Workaround: none — upgrade is required.

TLS certificate validation bypass in SOCKS5 ProxyAgent — CVE-2026-9697

GHSA-vmh5-mc38-953g · CWE-295

... (truncated)

Commits

• f9eba0a Bumped v7.28.0 (#5430)
• a027a4a Backport WebSocket maxPayloadSize fixes to v7.x (#5423)
• 8cb10f9 websocket: limit the number of fragments in a message
• 04201f8 fix: honor requestTls when proxy is SOCKS5
• fcd642f fix(socks5): preserve dispatch backpressure return value (#5166)
• bc98c97 fix(socks5): use configured connector in Socks5ProxyAgent (#5168)
• 9e1c743 fix(socks5): encode embedded IPv4 tails in IPv6 literals correctly (#5099)
• 376c8be fix(socks5): enforce authenticated state before CONNECT (#5097)
• 3805b8f fix(socks5-proxy-agent): use per-origin pools to prevent cross-origin routing...
• 85a2405 fix(cache): trim qualified field names
• Additional commits viewable in compare view

Updates @cloudflare/vitest-pool-workers from 0.12.5 to 0.18.7

Release notes

Sourced from @​cloudflare/vitest-pool-workers's releases.

@​cloudflare/vitest-pool-workers@​0.18.7

Patch Changes

• #14713 de34449 Thanks @​allocsys! - Fix a non-ASCII path failure during the Miniflare WebSocket handshake: the MF-Vitest-Worker-Data header embedded the raw process.cwd() value, which threw a Latin-1/ASCII header encoding error when the workspace path contained non-ASCII characters (e.g. CJK characters) on Windows. The value is now percent-encoded on write and decoded on read, matching the fix applied to the module fallback redirect response.

• #14713 de34449 Thanks @​allocsys! - Fix a runtime start-up failure ("No such module "cloudflare:test-internal"") when the project workspace path contains non-ASCII characters (e.g. CJK characters) on Windows. The module fallback service's redirect response set the target file path directly as an HTTP Location header value, but headers are restricted to the Latin-1/ASCII byte range, so any non-ASCII byte in the path caused header construction to throw. Such paths are now percent-encoded (and tagged with a sentinel prefix) before being used as a header value, and decoded again only for the values we encoded, so the round-trip is unambiguous and a workspace path containing a literal % is left untouched instead of being mis-decoded.

• #14739 5eac99e Thanks @​Ankcorn! - Support testing Streaming Tail Workers in Vitest Pool Workers.

• #14763 538e867 Thanks @​gianghungtien! - Treat webSocketMessage(), webSocketClose() and webSocketError() as optional Durable Object handlers

  The pool wraps each Durable Object class and installs a prototype method for every default handler before user code is loaded, so workerd always sees a handler and always dispatches. When the wrapped class didn't actually define one, the wrapper threw <ClassName> exported by <path> does not define a `webSocketClose()` method, even though deployed Workers silently ignore these events for classes that omit them. A hibernatable Durable Object defining only webSocketMessage() would log an uncaught TypeError on every close.

  These three handlers now no-op when absent, matching deployed behaviour. alarm() is unchanged and still reports a missing handler, since workerd rejects setAlarm() up front on a class without one.

• Updated dependencies [42af66d, a0a091b, f03b108, deae171, 0df3d43, d83a476, 4e92e32, d1d6945, 4815711, a0c8bb1, a50f73a, 2b390d7, c82d96b, 34430b3, f75ae5d]:

  • miniflare@4.20260721.0
  • wrangler@4.113.0

@​cloudflare/vitest-pool-workers@​0.18.6

Patch Changes

• #14678 4e62bba Thanks @​apeacock1991! - Fix test runs hanging after a Durable Object logs and rejects blockConcurrencyWhile()

  Console messages emitted from another Durable Object are now buffered until execution returns to the test runner, avoiding I/O that cannot complete after the object's input gate breaks.

• Updated dependencies [34e696d, d39ae01, 3de70df, c79504f, 9f04a7e, 9f04a7e, cb30df3, cb6c3f9, c7dbe1a, 3f3afbb, e6fbc4e, 4e1a7a7, 9f04a7e]:

  • miniflare@4.20260714.0
  • wrangler@4.112.0

@​cloudflare/vitest-pool-workers@​0.18.5

Patch Changes

• Updated dependencies [7692a61, ed33326, 018574b, eb99ab1, cdf3148, 7692a61, 7692a61, 3015320, 899c297, 9da77ac, 317ce1f]:
  • miniflare@4.20260710.0
  • wrangler@4.111.0

@​cloudflare/vitest-pool-workers@​0.18.4

Patch Changes

• #14535 1b965c5 Thanks @​Naapperas! - Support dynamic retry delays for Workflow steps in local dev

  A step's retries.delay can now be a function that computes the delay per failed attempt, in addition to a static duration. The function receives { ctx, error } and returns a delay (a number of milliseconds or a duration string like "30 seconds"), and its result is fed into the configured backoff.

  await step.do(
    "call flaky API",
    {
      retries: {
        limit: 5,

... (truncated)

Changelog

Sourced from @​cloudflare/vitest-pool-workers's changelog.

0.18.7

Patch Changes

• #14713 de34449 Thanks @​allocsys! - Fix a non-ASCII path failure during the Miniflare WebSocket handshake: the MF-Vitest-Worker-Data header embedded the raw process.cwd() value, which threw a Latin-1/ASCII header encoding error when the workspace path contained non-ASCII characters (e.g. CJK characters) on Windows. The value is now percent-encoded on write and decoded on read, matching the fix applied to the module fallback redirect response.

• #14713 de34449 Thanks @​allocsys! - Fix a runtime start-up failure ("No such module "cloudflare:test-internal"") when the project workspace path contains non-ASCII characters (e.g. CJK characters) on Windows. The module fallback service's redirect response set the target file path directly as an HTTP Location header value, but headers are restricted to the Latin-1/ASCII byte range, so any non-ASCII byte in the path caused header construction to throw. Such paths are now percent-encoded (and tagged with a sentinel prefix) before being used as a header value, and decoded again only for the values we encoded, so the round-trip is unambiguous and a workspace path containing a literal % is left untouched instead of being mis-decoded.

• #14739 5eac99e Thanks @​Ankcorn! - Support testing Streaming Tail Workers in Vitest Pool Workers.

• #14763 538e867 Thanks @​gianghungtien! - Treat webSocketMessage(), webSocketClose() and webSocketError() as optional Durable Object handlers

  The pool wraps each Durable Object class and installs a prototype method for every default handler before user code is loaded, so workerd always sees a handler and always dispatches. When the wrapped class didn't actually define one, the wrapper threw <ClassName> exported by <path> does not define a `webSocketClose()` method, even though deployed Workers silently ignore these events for classes that omit them. A hibernatable Durable Object defining only webSocketMessage() would log an uncaught TypeError on every close.

  These three handlers now no-op when absent, matching deployed behaviour. alarm() is unchanged and still reports a missing handler, since workerd rejects setAlarm() up front on a class without one.

• Updated dependencies [42af66d, a0a091b, f03b108, deae171, 0df3d43, d83a476, 4e92e32, d1d6945, 4815711, a0c8bb1, a50f73a, 2b390d7, c82d96b, 34430b3, f75ae5d]:

  • miniflare@4.20260721.0
  • wrangler@4.113.0

0.18.6

Patch Changes

• #14678 4e62bba Thanks @​apeacock1991! - Fix test runs hanging after a Durable Object logs and rejects blockConcurrencyWhile()

  Console messages emitted from another Durable Object are now buffered until execution returns to the test runner, avoiding I/O that cannot complete after the object's input gate breaks.

• Updated dependencies [34e696d, d39ae01, 3de70df, c79504f, 9f04a7e, 9f04a7e, cb30df3, cb6c3f9, c7dbe1a, 3f3afbb, e6fbc4e, 4e1a7a7, 9f04a7e]:

  • miniflare@4.20260714.0
  • wrangler@4.112.0

0.18.5

Patch Changes

• Updated dependencies [7692a61, ed33326, 018574b, eb99ab1, cdf3148, 7692a61, 7692a61, 3015320, 899c297, 9da77ac, 317ce1f]:
  • miniflare@4.20260710.0
  • wrangler@4.111.0

0.18.4

Patch Changes

• #14535 1b965c5 Thanks @​Naapperas! - Support dynamic retry delays for Workflow steps in local dev

  A step's retries.delay can now be a function that computes the delay per failed attempt, in addition to a static duration. The function receives { ctx, error } and returns a delay (a number of milliseconds or a duration string like "30 seconds"), and its result is fed into the configured backoff.

  await step.do(

... (truncated)

Commits

• 88f42e7 Version Packages (#14740)
• 0df3d43 [wrangler] Extract remote bindings into a package (#14720)
• 538e867 [vitest-pool-workers] Treat webSocket* as optional Durable Object handlers ...
• de34449 [vitest-pool-workers] Fix module fallback failure on non-ASCII workspace path...
• 5eac99e [vitest-pool-workers] Support streaming tails to the main worker (#14739)
• 82918ad Version Packages (#14699)
• 4e62bba fix(vitest-pool-workers): avoid hanging on broken DO logs (#14678)
• b449bb0 Version Packages (#14644)
• eced610 Version Packages (#14628)
• 774c09e Version Packages (#14603)
• Additional commits viewable in compare view