vercel/turbo

Why this matters

Turborepo is the canonical “modern, well-tooled, dual-language Rust+TS monorepo” — exactly the audience alint’s monorepo bundles target. A Rust workspace (crates/turbo*, 61 crates) plus a pnpm workspace (packages/*, apps/*, examples/*, 17 first-party packages plus 30 examples), with all language-level lints delegated to the canonical per-ecosystem tools — cargo fmt / cargo clippy / cargo deny for Rust, oxlint / oxfmt / taplo / attw for TS — and turborepo itself + a husky pre-push hook orchestrating them. There is no xtask/ crate, no .changeset/ directory, no hack/verify-*.sh pipeline.

That means the structural-validation surface area is small and almost entirely about monorepo conventions rather than language semantics — exactly the layer where alint adds a structural floor that doesn’t exist in the existing tooling stack.

Headline catch

Kubernetes has 50 hand-rolled verify scripts → alint replaces 17. Turborepo has zero hand-rolled verify scripts (everything delegated to per-language tools) → alint adds ~22 structural gates that don’t exist today at all.

Even a project with Vercel-grade tooling has structural drift that no per-language linter catches.

The findings on the actual repo are real and actionable:

• 60 of 61 crates are missing the publish = false guard — one cargo publish --workspace from a maintainer’s machine would otherwise leak internal-only crates to crates.io
• 8 of 17 packages lack a per-package LICENSE (problematic for npm pack since the repo-root LICENSE doesn’t auto-include in tarballs)
• 9 of 52 crates are missing READMEs
• 7 crates have a directory name that doesn’t match the published crate name (crates/turborepo-globwalk is published as globwalk, crates/turborepo-paths as turbopath, etc.) — intentional namespacing drift, but new drift is almost always a mistake
• One example (with-microfrontends) is silently skipped from the check-examples.ts sandbox runs because it lacks meta.json; with-nextjs is missing turbo.json

These are conventions enforced today only by package.json / Cargo.toml review etiquette. None of them surface from running cargo clippy, oxlint, taplo, or attw.

Where alint earns its keep here

• One declarative file (~22 structural gates) replaces what’s currently spread across pre-push shell + lint.yml jobs + a TypeScript runner. for_each_dir over the 61 crates + 17 packages + 30 examples covers the per-package layout, manifest fields, license presence, and meta.json/turbo.json shape in tens of milliseconds.
• 7 command: rule shell-outs to the canonical per-language tools (cargo fmt / cargo clippy / cargo deny / oxlint / oxfmt / taplo / shellcheck) — alint isn’t faster than the underlying tools, but running them all in parallel under one alint check shaves wall time vs. the sequential pre-push chain.
• dir_name_matches_field is the v0.10+ headline demand-driver here. Both Cargo and npm let the directory name diverge from the published name; turbo has 7 such crates and 7 such packages that are intentional, but new drift is almost always a mistake. The v0.10+ design needs a paths.exclude: or allow_drift: knob — turbo is the canonical “expected drift, allowlist 7” demonstration.
• json_schema_passes is the v0.10+ second-most-load-bearing missing primitive. Turborepo ships its own schema (docs/public/schema.json) and expects users to validate turbo.json against it via the $schema field. tsconfig.json, oxlint configs, etc. all publishing schemas — broad demand even if per-repo confirmations are modest.
• The alint pr-diff-check sibling-mode candidate — Turborepo has a hand-rolled “release PRs may only touch version.txt / package.json / Cargo.toml / Cargo.lock / CHANGELOG / pnpm-lock.yaml” guard via gh api repos/.../pulls/N/files. A separate-binary alint pr-diff-check mode is a v0.10+ design candidate; doesn’t fit the alint check (repo-state-only) model.

Future story angles

• The kubernetes ↔ turbo contrast is the launch hook — 50 hand- rolled scripts to alint replacing 17 on one end; zero hand-rolled scripts to alint adding 22 structural gates that don’t exist on the other. Use this on alint.org/examples as evidence that monorepo conventions are under-checked even at top-tier-tooling repos.
• Hybrid-monorepo launch narrative — pair turbo with next.js (the dual pnpm + Cargo mega-monorepo) for a “modern monorepo tier” launch tile pair. Same Vercel ecosystem; very different scales (turbo: 17 packages + 61 crates; next.js: 19 packages + 63 crates + 30 workflows); both have structural drift no per-language linter catches.
• dir_name_matches_field v0.10+ design narrative — turbo is the demand source AND the expected-drift example. When the rule kind ships, the launch-post writes itself: “turbo has 7 crates with intentional name/dir drift; alint catches the new drift while letting the historical drift be allow-listed.”
• scope_filter evolution refactor — crates//packages// examples/ triad as named scopes; cuts ~20 lines and clarifies per-example rule intent. Canonical demonstration of the v0.9.17 named-scope payoff for triple-subtree monorepos.