chore: use pnpm and harden security#221
Merged
Merged
Conversation
246b97c to
41292ce
Compare
staaldraad
approved these changes
May 12, 2026
mandarini
added a commit
that referenced
this pull request
Jun 9, 2026
## Summary - Re-adds an explicit npm install before the publish step so the workflow runs on **npm 11.5.2** (the version that supports OIDC trusted-publisher exchange) instead of Node 22's bundled npm 10.9.x. - Adds a multi-line comment in the workflow explaining the dependency, so the line is not deleted again. ## Background The release pipeline has been silently broken since #221 (May 12). That PR removed `corepack enable npm && corepack prepare npm@11 --activate` with the claim "no longer needed", but `actions/setup-node@v6.4.0` does **not** upgrade npm — Node 22 ships with npm 10.9.x. npm only learned how to do OIDC trusted-publisher exchange in **11.5.1**. On npm 10, `npm publish --provenance` signs the provenance via sigstore (which works because it uses the GitHub OIDC token directly), then sends the bogus `.npmrc` placeholder `XXXXX-XXXXX-XXXXX-XXXXX` as the bearer token to the registry. The registry returns `404 Not Found - PUT https://registry.npmjs.org/@supabase%2fssr` (npm registry returns 404 for unauthenticated PUTs to avoid leaking which packages exist). The repo has **no `NPM_TOKEN` secret** configured anywhere — by design, per #221's hardening posture. Auth is OIDC-only via the trusted-publisher binding on npmjs.com. Re-adding a token would weaken that posture, so the fix is to ensure the npm CLI is new enough to use OIDC. ## Why this went undetected Between #221's merge (May 12) and #240's merge (June 4), every release run was for a `chore: update @supabase/supabase-js` commit. Those have no pending release-please PR, so the workflow's version-determination step took the `skip=true` branch and never attempted to publish. The first publish attempt on npm 10 — #240's merge on June 4 — failed with E404, as did the two release runs after it (#245, #244). Failed runs: - [27136548767](https://github.com/supabase/ssr/actions/runs/27136548767) — PR #244 ("release 0.11.0") merge, tried to publish `0.12.0-rc.118` - [27008002122](https://github.com/supabase/ssr/actions/runs/27008002122) — PR #245 merge - [26949675017](https://github.com/supabase/ssr/actions/runs/26949675017) — PR #240 merge, tried to publish `0.11.0-rc.117` Last successful publish: `v0.10.3` on May 7 (run [25509681243](https://github.com/supabase/ssr/actions/runs/25509681243)), which ran on npm 11 via the corepack line that #221 removed. Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Migrates the repo to pnpm and applies the low-risk subset of supply-chain hardening from the playbook informed by the TanStack/router compromise (May 2026) and Adnan Khan's GitHub Actions cache poisoning research. End users of
@supabase/ssrare unaffected: same tarball, samenpm publish --provenanceflow, same trusted-publisher binding.Package manager
packageManagertopnpm@11.1.0with a sha512 hash inpackage.json. A tampered pnpm binary served by a compromised mirror fails the hash check instead of running.package-lock.jsontopnpm-lock.yaml(3589 lines removed, 3213 added). Lockfile contains only registry resolutions (nogithub:,git+,file:refs).Supply-chain hardening (
pnpm-workspace.yaml)New file with three guards aimed at the TanStack class of attacks:
minimumReleaseAge: 10080blocks installs of any version published in the last 7 days. The TanStack malicious versions were pulled within hours, so this single setting would have blocked the install path.minimumReleaseAgeExclude: ['@supabase/*']lets theupdate-supabase-js.ymlworkflow bypass the quarantine for first-party releases.blockExoticSubdeps: truerefuses non-registry refs in transitive deps. This kills theoptionalDependencies: { foo: "github:owner/repo#sha" }vector TanStack used.allowBuilds: { esbuild: true }is the explicit install-script allowlist (default deny). Only esbuild is permitted to run lifecycle scripts, since vitest depends on it.Workflow hardening
Applied across all four workflows (
ci.yml,release.yml,update-supabase-js.yml,conventional-commits.yml):persist-credentials: falseon everyactions/checkout. Default behavior writes a write-capable token to.git/configon disk, where any later step (including a compromised dependency) can read it.pnpm install --frozen-lockfile,pnpm build,pnpm test,pnpm exec prettier,pnpm exec eslint.pnpm/action-setup@739bfe42ca9233c5e6aca07c1a25a9d34aca49b0(v6.0.7), pinned by commit SHA. v6 reads thepackageManagerfield automatically so noversion:arg is needed.update-supabase-js.ymlcheckout SHA from v4.3.0 to v6.0.1 so dependabot only tracks one version.corepack enable npm && corepack prepare npm@11line fromrelease.yml(no longer needed).# v4.4.0comment next to thegoogleapis/release-please-actionSHA pin for readability.Cache poisoning fix
Removed
cache: pnpmfromrelease.ymlandupdate-supabase-js.yml. Per Adnan Khan's research, caching in release workflows is exploitable: a compromised dep that runs on a main-context job can dump the Actions runtime token from runner memory and poison cache entries that a later release run will restore, replacingpackage.jsonwith one carrying apreinstallpayload. Withid-token: writein scope, that payload can mint a publish token.ci.ymlretainscache: pnpmbecause cache poisoning there only breaks CI, not releases.Other changes
.github/dependabot.yml(new): weeklygithub-actionsecosystem updates so SHA-pinned actions stay current..github/CODEOWNERS(new): blanket ownership across the repo so workflow and release-config edits require explicit review.README.md: install snippet now shows npm, pnpm, yarn, and bun side by side instead of pushing consumers toward one manager.src/createBrowserClient.tsandsrc/createServerClient.ts: prettier reformatted these (whitespace only, no semantic change) after the prettier version resolved by pnpm bumped from the previous lockfile pin to 3.8.3. No runtime difference.