Status as of the original triage: 21 advisories -> 3. This document explains what was fixed, what was deliberately left, and why.
Since then: 6 Next.js advisories -> 0, fixed by upgrading 16.3.6 to 16.3.8. See Next.js below. The standing accepted count is still 3.
Re-checked 2026-10-12 against the GitHub advisory database and the npm registry: the three open Dependabot alerts are the same three, pnpm audit still reports exactly three advisories and no more, and none of them has become fixable in the meantime. Each entry under Not fixed carries its own re-verified date and evidence, and Re-checked 2026-10-12 records why no fix is reachable today. That pass also corrected the exposure of decode-uri-component: it is runtime-reachable, not build-toolchain-only.
Severity labels come from the advisory database. Exposure is a separate question and is recorded per finding below, because most of the original count was toolchain code that never ships.
Reproduce with:
pnpm audit
pnpm audit --jsongh api repos/usequark/quark/dependabot/alerts returns 403 from CI, so pnpm audit is the source of truth.
| Package | Advisories | Severity | Fix | Exposure |
|---|---|---|---|---|
nodemailer 7.0.13 | 13 | high x3, moderate x9, low x1 | Peer edge removed | none - dead code |
qs 6.15.3 | 2 | moderate |
overrides, inside stripe's own ^6.11.0 |
| production |
shell-quote 1.10.0 | 1 | critical | lockfile refresh | dev only |
source-map-js 1.2.1 | 1 | high | lockfile refresh | build only |
uuid 7.0.3 | 1 | moderate | overrides, justified below | build only |
Six advisories landed against [email protected] after the triage above. All six are patched in 16.3.8, so the fix is a version bump rather than an allowlist entry:
| Advisory | Severity | Issue |
|---|---|---|
GHSA-cjq9-62q9-8jv4 | high | SSRF in Image Optimization |
GHSA-mcj8-r9mp-w47p | moderate | SSG/ISR cache poisoning, cross-user content substitution and DoS |
GHSA-4jqv-mc3x-m676 | moderate | SSG/ISR cache poisoning in self-hosted apps |
GHSA-3w37-wq28-93x7 | moderate | Pending use cache fill can leak Draft Mode content |
GHSA-f87g-xv8r-7p7x | moderate | Metadata image route disclosure via dynamicParams bypass |
GHSA-39w2-rjm5-chcv | low | Dev server MCP endpoint disclosure |
These are not the accepted three. Two are production-reachable (SSRF in image optimization, ISR cache poisoning), so suppressing them would have been the wrong call regardless of the gate's rules. apps/web and packages/ui both move to 16.3.8, and sync-templates propagates the bump to scaffolded projects so new apps do not ship the vulnerable pin.
This was the largest single block in the audit and the least dangerous of them.
packages/core declares nodemailer: ^10.0.10 and resolves 10.0.10, which is past every advisory's patched range. The vulnerable copy was a second, older 7.0.13 pulled in as an auto-installed optional peer:
apps/web -> [email protected] -> @auth/[email protected]
peerDependencies: { nodemailer: "^7.0.7 || ^8.0.5" }
peerDependenciesMeta: { nodemailer: { optional: true } }Two facts make this unfixable by upgrading:
>=10.0.6. No version satisfying ^7.0.7 || ^8.0.5 is patched, so there is no in-range upgrade.@auth/[email protected] is the newest published release and @auth/[email protected] is also current. Neither has widened the range.Rather than force nodemailer two majors past a range its author chose, we remove the peer edge entirely, which deletes 7.0.13 from the tree:
"next-auth>nodemailer": "-"
"@auth/core>nodemailer": "-"This is safe because the provider is never used:
apps/web/src/lib/auth.js configures CredentialsProvider, GithubProvider and GoogleProvider only.@auth/core's index never imports nodemailer statically. Only @auth/core/providers/nodemailer.js does, and nothing imports that module.@usequark/quark-core (src/email.js), which depends on nodemailer ^10.0.10 directly.Deleting the peer edge means next-auth/providers/nodemailer stops resolving. Left alone, that failure is late and opaque: the provider does import { createTransport } from "nodemailer" at module scope, so it surfaces as ERR_MODULE_NOT_FOUND on the next server start, with nothing pointing at the override.
scripts/check-standards.mjs now fails the build if any source file imports that provider, and names the fix in the error. It runs in CI via pnpm standards, and covers the monorepo and the scaffold template.
The guard matches every spelling that resolves the module - static default, bare side-effect, re-export, await import(...), import(...).then(...) and require(...), including single quotes, template literals and a specifier wrapped onto its own line. It deliberately does not fire on other next-auth/providers/* imports, on a bare nodemailer import, on an identifier that merely contains import or require, or on a comment that mentions the path. scripts/check-standards.test.mjs has 20 cases covering both directions, because a guard that misses a form reads as protection while letting the breakage through.
[email protected] declares qs: ^6.11.0. Both advisories need >=6.16.0, which is inside that range, so the override does not violate anything stripe declared. pnpm update cannot make this move because qs is not a direct dependency of any workspace package.
qs@>=6.14.2 <=6.15.3: ">=6.16.0"Both were stale lockfile entries that a plain pnpm update -r moved forward, staying inside their parents' declared ranges. No override was added for either. The critical severity on shell-quote was misleading: it is a root devDependency reached through @changesets/cli -> launch-editor, and the advisory is command injection in a local editor-launcher on a developer machine, not in anything deployed.
[email protected] declares uuid: ^7.0.3; the advisory needs >=11.1.1. This is the one override that leaves a declared range, so it was checked rather than assumed:
xcode only calls uuid.v4(). That is the sole call site, at lib/pbxProject.js:90, inside generateUuid().expo prebuild --platform ios produced apps/mobile/ios/Quark.xcodeproj/project.pbxproj, the audited chain was driven directly: generateUuid() is the function that would break if the override were wrong, and it returns a correctly formatted 24-character uppercase hex ID with uuid 14.0.2 resolved.pbxProject = require('.../xcode/lib/pbxProject.js') # requires uuid at module scope
proj = new pbxProject('ios/Quark.xcodeproj/project.pbxproj')
proj.parseSync() # 32 UUIDs parsed
proj.generateUuid() -> '95C8FFC7A2D447F4B866B875' # 24-hex uppercase: true@expo/config-plugins when generating native projects. Nothing about it ships in a JS bundle.These three remain. None has a fix we can take without breaking something, and all three sit in apps/mobile's Expo dependency graph.
Two of the three are build-toolchain only - node-forge and braces never enter a JS bundle. The third, decode-uri-component, is the exception: it is reachable at app runtime, as corrected on 2026-10-12 after review. The original triage's blanket claim that none of them is shipped application code was wrong for that one, so exposure is now stated per entry.
Re-verified 2026-10-12. The resolved version and the module are unchanged from the original triage for all three, and no upstream package has published a patched release that a bump or an override could reach.
apps/mobile -> [email protected] -> [email protected] -> decode-uri-component (re-checked with pnpm why decode-uri-component; the same chain repeats through @expo/[email protected] -> @expo/[email protected])>=0.5.0. [email protected] declares ^0.2.2 and is CommonJS; it does const decodeComponent = require('decode-uri-component') and then calls it as a function. [email protected] is ESM-only, so under require(esm) it returns a namespace object. Verified directly: the direct call fails with TypeError: decodeComponent is not a function, while decodeComponent.default(...) works. Forcing the upgrade would break query-string at runtime.[email protected] into [email protected]'s tree reproduces the failure on Node v24.21.0 for parse, parseUrl, pick and exclude. [email protected] is "type": "module" with an exports map and no CommonJS entry point, so require() returns { __esModule, default }. The message name differs from the original note (m is the minified binding) but the cause is the same.[email protected] - the newest stable release, on both the latest and sdk-57 dist-tags - still pins query-string: ^7.1.3. [email protected] also still depends on decode-uri-component@^0.4.1, which is inside the vulnerable <=0.4.2 range, so even a 7 -> 8 move changes nothing. Expo SDK 58 exists only on the next and canary dist-tags ([email protected], [email protected]), so there is no stable release to move to.[email protected] is the first version that depends on decode-uri-component@^0.5.0, but it is also ESM-only, so require("query-string") returns { __esModule, default } and qs.parse(...) throws TypeError: qs.parse is not a function. Verified in a scratch install on the same Node. It is additionally a two-major jump over what expo-router declares, on a path that runs inside the app bundle - the guardrails rule that out.query-string is in the shipped mobile bundle, and it parses incoming URL query strings while the app runs. The chain, read off node_modules: So a deep-link query string reaches the vulnerable decoder on device. apps/mobile/app.json declares "scheme": "quark", so quark://...?q=... links are handled by the app. (Establishment is module-graph analysis, not an emitted bundle: expo export cannot currently produce one - see "Known unrelated breakage" below.)apps/mobile package.json "main": "expo-router/entry"
-> expo-router/entry.js -> entry-classic -> build/qualified-entry
-> build/ExpoRoot.js:135 <NavigationContainer linking={store.linking} ... />
-> build/fork/NavigationContainer.js:12 require("./useLinking") (native: useLinking.native)
-> build/fork/useLinking.native.js:46 Linking.addEventListener('url', ...)
build/fork/useLinking.native.js:115 getStateFromPath(path, config)
-> build/react-navigation/core/getStateFromPath.js:41 require("query-string")
build/react-navigation/core/getStateFromPath.js:499 queryString.parse(query)
-> [email protected]/index.js:3 require('decode-uri-component')
[email protected]/index.js:233 decodeComponent(value) <- 0.2.2[email protected] is ESM-only, so under [email protected]'s require('decode-uri-component') decodeComponent is a namespace object and every call throws TypeError: decodeComponent is not a function - i.e. the app breaks on every deep link that carries a query parameter, including well-formed ones. Overriding [email protected] instead breaks it at module scope. A crash on all links is strictly worse than a crash on a pathological one, so the fix is not an improvement and is not taken. Node v24.21.0, reproduced.query-string bumping to a version that drops the dependency, or a patched decode-uri-component release that keeps a CommonJS entry point. Both are tracked as the follow-up that would actually retire this advisory.apps/mobile -> [email protected] -> @expo/[email protected] -> [email protected] (also via @expo/[email protected]) - both re-confirmed with pnpm why node-forge.<0.0.0. There is no fixed release. 1.4.0 is the newest version published.first_patched_version: null and vulnerable_version_range: <= 1.4.0. The registry's version list ends at 1.4.0 and no dist-tag carries a newer or prerelease build, so there is nothing to upgrade to and nothing to pin an override to.[email protected]'s bin is bin/cli, so @expo/cli is reached through the CLI binary, and expo's own src/build never imports @expo/cli. Inside @expo/[email protected], node-forge is imported only by src/utils/codesigning.js and src/run/ios/codeSigning/Security.js - iOS code signing. The app bundle follows imports from expo-router/entry and never reaches either.@expo/cli moves off [email protected]. The newest @expo/[email protected] still declares node-forge: ^1.3.3. Do not force an override: there is no version to force it to.<0.0.0. No fixed release exists; 3.0.3 is the newest version.first_patched_version: null and vulnerable_version_range: <= 3.0.3, and the registry's version list ends at 3.0.3. micromatch's newest release, 4.0.8, declares braces: ^3.0.3, so even the newest parent resolves the vulnerable version. Nothing to upgrade to.pnpm why braces) - and there is more than one. The advisory names one chain, apps/mobile -> [email protected] -> @expo/[email protected] -> @expo/[email protected] -> [email protected] -> [email protected], but pnpm why braces shows braces entering the tree through three independent routes:@expo/cli -> @expo/[email protected] -> [email protected] -> braces@expo/cli -> @expo/[email protected] -> [email protected] and ... -> [email protected] -> [email protected] -> braces (and again via @expo/cli -> @expo/[email protected] -> @expo/[email protected])[email protected] -> @react-native/[email protected] -> [email protected] -> [email protected] -> [email protected] -> braces@expo/metro-file-map bump would not retire this advisory: @expo/[email protected] does drop micromatch from its own manifest, but that only removes route 1. Routes 2 and 3 resolve [email protected] regardless, so pnpm audit keeps reporting it. Bumping it would be churn with no security outcome, so it was not done.micromatch matches file paths against developer-authored glob patterns at bundle time. Reaching the DoS needs a pathological pattern authored in the repo, not attacker-controlled input. Re-checked 2026-10-12 rather than assumed: every route into micromatch runs through metro-file-map, which is Metro's file crawler and runs in Node during expo start/expo export. [email protected]'s own src/build never imports metro-file-map or micromatch, and nothing in apps/mobile's source does either, so it does not enter the app bundle. Unlike decode-uri-component, this one is genuinely not shipped.metro-file-map stops depending on [email protected] on every route above, or if micromatch moves to a braces range that admits a fixed version.Checked against the GitHub advisory database (gh api repos/usequark/quark/dependabot/alerts) and the npm registry, per advisory:
| Advisory | Newest published | GitHub first_patched_version | Reachable by | Verdict |
|---|---|---|---|---|
node-forge GHSA-86w9-cpqp-85rv | 1.4.0 | null | nothing - no patched release exists | still accept, build-only |
braces GHSA-vfj7-8cjw-p6xm | 3.0.3 | null | nothing - no patched release exists | still accept, build-only |
decode-uri-component GHSA-vcc3-ghjq-m6fr | 0.5.0 | 0.5.0 | only by overriding past [email protected] | still accept, runtime-reachable |
A lockfile refresh was considered for all three and cannot move any of them:
node-forge and braces have no newer release at all, so any resolution lands on the same affected version.[email protected] declares decode-uri-component: ^0.2.2, and a caret range on a 0.x version pins the minor (>=0.2.2 <0.3.0). 0.2.2 is already the newest version that range can resolve, so a refresh keeps 0.2.2 and only a deliberate override could change it.The only edits that could take the decode-uri-component fix - overriding decode-uri-component to 0.5.0, or query-string to 9.x - both break the CommonJS caller, and the second is a two-major jump on a path that runs inside the app bundle. Both were tested rather than assumed. No package.json or lockfile change was made.
One correction made during this pass, after review: the original triage recorded all three as build-toolchain only. That holds for node-forge and braces, but decode-uri-component is runtime-reachable — expo-router's linking stack calls queryString.parse on incoming deep-link query strings, so a malformed one reaches the vulnerable decoder on device. The verdict is unchanged (the fix would break every deep link, not just the pathological one), but the exposure is recorded as it is rather than as the old blanket claim described it.
Run pnpm audit:check. If any of the three remaining advisories disappears, it moved upstream and this document should be updated - the check prints a notice naming the entry to retire. Note that pnpm audit reads the lockfile, so a stale .pnpm directory can make a resolved tree look unfixed - rm -rf node_modules && pnpm install if a count looks stale.
A refresh of the lockfile on its own will not retire any of the three, for the reasons listed under Re-checked 2026-10-12. The signal to watch for is an upstream release, not a lockfile resolution.
pnpm audit:check runs scripts/check-audit.mjs, which CI invokes on every push and pull request. The three accepted advisory IDs are an allowlist in that script, so it passes today and turns red when a fourth appears.
A bare pnpm audit step could not do this: it exits non-zero for any advisory, so it would fail forever on the three above and get ignored. The allowlist is keyed on the GitHub advisory ID because that is stable across advisory metadata edits in a way the npm advisory ID is not.
The script treats "the audit did not run" as a failure rather than as a clean tree. pnpm audit reports its own failures as JSON on stdout with exit 1 - for example {"error":{"code":"ERR_PNPM_AUDIT_NO_LOCKFILE"}} - which has no advisories key at all. Reading that as zero advisories turns a missing lockfile or an unreachable registry into a green build that verified nothing, which is the one failure mode a security gate must not have.
It fails closed on unrecognised report shapes for the same reason: an advisories key that is present but null, or any non-object, is rejected rather than defaulted to an empty set. So is an advisory with no usable github_advisory_id - a legacy or CVE-only record with a null ID can never match the allowlist, so dropping it would report a clean tree while a vulnerability is present. An advisory the check cannot identify is one it cannot clear. scripts/check-audit.test.mjs (17 tests) covers both directions, including these cases.
Found while verifying the uuid override above. Not caused by any override in this document - it reproduces on origin/main with a clean lockfile.
expo export fails:
iOS Bundling failed
Error: Cannot find module '.../react-native/rn-get-polyfills'Cause: @expo/[email protected] calls require(<react-native>/rn-get-polyfills)() in its Metro getPolyfills hook. react-native shipped that file through 0.86 and dropped it in 0.87 - the published [email protected] tarball contains zero occurrences of it. apps/mobile pins react-native: 0.87.1, but Expo SDK 57 pins 0.86.3, which is what npx expo install --check reports as expected.
Narrower than it looks: expo prebuild works fine on 0.87.1, because it never reaches the Metro polyfill hook. Only the bundler path breaks.
Downgrading to 0.86.3 was tested and does fix it - expo export produced a complete iOS bundle (entry-8badf830eadac4e04fbda158e097aeea.hbc, 2.9MB). It was left undone here because it is a dependency decision outside the scope of an advisory audit: 0.86.3 is what Expo asks for, but taking it means reverting a Dependabot bump and re-testing the mobile app on an older RN. expo install --check lists 11 further packages that want the same treatment.
Related: pnpm build:mobile invokes a script the mobile package does not have (pnpm --filter @usequark/quark-mobile export -> "None of the selected packages has an 'export' script"), so it has never worked. Also pre-existing.
auditConfig.ignoreCves or audit.level change. The three remaining advisories are reported honestly rather than silenced. Those pnpm settings mute the audit database, which would hide the accepted three and anything new; the allowlist in scripts/check-audit.mjs is code, reviewable in a diff, instead.pnpm audit, and pnpm audit is already enforced in CI.