Skip to content

chore(deps): bump undici from 6.29.0 to 6.29.1 - #18

Merged
twk3 merged 1 commit into
mainfrom
claude/vibrant-shannon-7e56v9
Oct 27, 2026
Merged

twk3 merged 1 commit into
mainfrom
claude/vibrant-shannon-7e56v9

Conversation

@maxigimenez

@maxigimenez maxigimenez commented Oct 27, 2026 •

Copy link
Copy Markdown

Owner: @maxigimenez
Linear: n/a — opened by the scheduled dependency security triage

What and why

undici 6.29.0 → 6.29.1 (high; transitive, via @actions/http-client under @actions/core).

GHSA-xjj6-f33q-929c / CVE-2026-109869, patched in 6.29.1. The advisory's own impact text:

undici's fetch() crashes the Node.js process when a malicious server returns a response with more than five content codings and closes the connection before sending the body. undici rejects such responses (a content-encoding limit), but it created the internal response stream before that check and returned without attaching an error listener or aborting the request. The early close then surfaces a content-length mismatch on that orphaned internal stream, which Node.js treats as an unhandled error and terminates the process. A remote, unauthenticated server can trigger this with a single response, even when the application correctly catches the fetch() rejection.

One bump closes five advisories. npm audit reports 0 vulnerabilities afterwards:

advisory CVE severity
GHSA-xjj6-f33q-929c CVE-2026-109869 high
GHSA-qrxm-fqhp-jwvm CVE-2026-87018 high — TLS 1.2 renegotiation crash
GHSA-2m5v-pch6-74hx CVE-2026-109383 high — Agent dropped function-valued TLS options such as checkServerIdentity
GHSA-4qv8-2p9j-2685 CVE-2026-109924 moderate — unbounded EventSource parser state
GHSA-9fm3-97px-rjf6 CVE-2026-109932 moderate — unbounded WebSocket PONG buffering

Vanta has no finding for this repo at all, so none of these has an SLA clock. It files CVE-2026-109869 against currents-playwright (alert 417) and playwright-last-failed (alert 86), both due late November, but nothing here. This came out of the lockfile sweep rather than the worklist — scanner coverage, not a sign the finding is minor.

Exposure

Runtime, and it ships. The lockfile stanza is not marked dev, and the chain starts at a root dependencies entry:

node_modules/undici — 6.29.0 (no "dev": true)
requesters: node_modules/@actions/http-client -> ^6.23.0  ←  @actions/core (root dependency)

This is the HTTP client the action uses to call the Currents API, and it is inlined into dist/index.js, which is what actually runs when a workflow references this action. So the vulnerable parser ships to every consumer of the action until this lands — unlike the sibling ide-extension case, where undici is dev-only.

That said, reachability is narrow: the action talks to the Currents API, not to arbitrary attacker-chosen hosts, so triggering it needs that endpoint (or something impersonating it) to return a malformed response.

The fix

lockfile-upgrade. @actions/http-client's ^6.23.0 already admits 6.29.1, so package.json is unchanged.

Two files move, and the second is required rather than incidental:

  1. package-lock.json — three lines in one stanza (version, resolved, integrity).
  2. dist/index.js + .map — regenerated with npm run package. The bundle inlines undici, so a lockfile-only change would leave the vulnerable code in what actually executes, and check-dist.yml would fail the PR for a stale bundle.

You can confirm the fix landed in the bundle — the patched form replaces reject(...) with this.abort(...):

this.abort(new Error(`too many content-encodings in response: ${codings.length}, maximum allowed is ${maxContentEncodings}`))

grep -c 'reject(new Error(too many content-encodings'on the newdist/index.js` returns 0.

On the lockfile diff being exactly 3 lines

npm update undici in this sandbox also strips "libc": ["glibc"] / ["musl"] from 10 dev-only optional binary stanzas — sandbox npm 10.9.4 writing metadata differently from CI. Those fields are the only discriminator between the gnu and musl variants of the same binary, so losing them lets an Alpine install pick the wrong one. I reverted that and applied the undici stanza by hand, then re-ran npm ci and rebuilt dist/ from that clean install, which is why the lockfile diff is 3 lines rather than ~35.

Breaking-change check

  • 6.29.1 — the five security fixes above, plus one functional change, nodejs/undici#6000 "fix(fetch): fix aborting Streams". No breaking-change or removal section.

Single patch bump within major 6. No patches/ directory and no postinstall, so the patch-package trap does not apply.

Verification — run on this branch head

command result
npm ci (from the hand-edited lockfile) clean, no peer warnings
installed version check node_modules/undici reports 6.29.1
npm run all (build → format-check → lint → package → test) pass, exit 0
npm test 26/26 tests pass, 1 suite
git diff --name-only package-lock.json, dist/index.js, dist/index.js.map — nothing else
npm audit 0 vulnerabilities

On dist/ reproducibility, since CI checks it byte-for-byte

check-dist.yml rebuilds with Node 24.x and fails if git diff dist/ is non-empty. This sandbox only has Node 22.22.0, so I verified the toolchain is deterministic across that gap before trusting the regenerated bundle:

On unmodified origin/main, npm ci && npm run package under Node 22.22.0 / ncc 0.44.1 reproduced the committed dist/ with 0 diff lines.

Since the committed bundle reproduces exactly here, a bundle generated here should likewise reproduce on CI's Node 24. If check-dist fails anyway, that is the Node-major difference and not the dependency change — the fix is to rerun npm run package on Node 24 and commit that.

CI coverage

Both relevant workflows trigger on pull_request with no paths: filter, so this diff runs them rather than silently skipping:

  • test.yml (build-test) — build and the Jest suite.
  • check-dist.yml — rebuilds dist/ and compares; paths-ignore is only **.md, which this diff does not match. This is the one that would have caught a lockfile-only fix.
  • codeql-analysis.yml — runs on pull_request to main.

Why this is Tier B and does not auto-merge

Three independent reasons, any one of which is enough:

  1. undici is on the sensitive-package list by name, whatever scope it resolves to.
  2. It is a runtime dependency that ships in the published action bundle — runtime entries never auto-merge, transitive or not.
  3. The diff is not lockfile-plus-one-manifest; it carries a regenerated bundle.

What a human should still check

  1. That check-dist passes on CI's Node 24. This is the one thing the sandbox could not fully prove — see the reproducibility note above for what to do if it does not.
  2. Whether consumers pin this action by tag or SHA, since the fix only reaches them when a release moves the tag they use. This routine does not create tags or releases.

Generated by Claude Code


View with [code]smith Autofix with [code]smith
Need help on this PR? Tag @codesmith-bot with what you need.

Closes GHSA-xjj6-f33q-929c / CVE-2026-109869 (high): undici's fetch() created
the internal response stream before rejecting a response with more than five
content codings, so a server that closed the connection early surfaced a
content-length mismatch on an orphaned stream and terminated the process.

undici arrives through @actions/http-client under @actions/core, whose
^6.23.0 range already admits 6.29.1, so package.json is unchanged. dist/ is
regenerated because the bundle inlines undici and check-dist compares it
against a fresh build.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_011ya3soxe5kkfQeEw42FD6u
@maxigimenez
maxigimenez requested a review from twk3 October 27, 2026 09:18
@baz-reviewer

baz-reviewer Bot commented Oct 27, 2026 •

Copy link
Copy Markdown

Review this PR on Baz

Baz Summary

Upgrade the bundled undici HTTP client from 6.29.0 to 6.29.1 to incorporate fixes for fetch process crashes, TLS option handling, EventSource memory growth, and WebSocket backpressure. Regenerate the shipped action bundle and source map so runtime consumers receive the patched implementation, while updating the lockfile to pin the secure dependency.

Topics

TopicDetails
Runtime security fixes Apply undici security and resilience fixes across fetch response handling, TLS configuration, EventSource parsing limits, and WebSocket write backpressure to protect runtime HTTP and streaming flows.
Modified files (1)
  • dist/index.js
Latest Contributors(2)
UserCommitDate
noreply@anthropic.comchore(deps): bump undi...October 27, 2026
agoldis@gmail.comfix: do not follow red...July 27, 2026
Dependency packaging Pin undici 6.29.1 and regenerate the distributable bundle artifacts so the published action contains the patched dependency rather than only updating development metadata.
Modified files (2)
  • dist/index.js.map
  • package-lock.json
Latest Contributors(2)
UserCommitDate
noreply@anthropic.comchore(deps): bump undi...October 27, 2026
gimenez.maxi@gmail.comfix: move brace-expans...October 06, 2026


Merger  Activate to get a short verdict whether this PR is good to go or not

Skills  Activate Skill Maintainer to keep your skills up to date

Planner  This PR would have been improved with Baz Planner - Try it now

Copy link
Copy Markdown
Author

CI has now answered the one open question in the description: check-dist passed on c746156.

That was the real risk here — the bundle was built in a sandbox on Node 22.22.0 while check-dist.yml rebuilds on Node 24.x and fails on any byte of difference. It reproduced, so the Node-major gap did not matter and the committed dist/ is what CI's own build produces.

All 6 checks green: check-dist, build, test, CodeQL, Analyze (javascript-typescript), Baz Reviewer.

That closes item 1 of "what a human should still check". Item 2 stands and is not something CI can answer: this fix only reaches consumers when a release moves the tag they reference — the action ships undici inlined in dist/index.js. This routine does not create tags or releases.

It does not auto-merge: runtime scope, ships in the published bundle, and undici is on the sensitive-package list by name — any one of those is disqualifying. Ready for a human.


Generated by Claude Code

@twk3
twk3 merged commit c50c569 into main Oct 27, 2026
6 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants