Commands

The root manifest carries a script per check, per build target and per filtered subset of each, and the subsets are where the count comes from: lint exists four times because it is crossed with two filters and a fix flag, and test and typecheck exist twice each for the same reason. A task runner that takes filters as arguments makes most of those spellings unnecessary.

This page is the inventory. It assumes the phases in order — a script only loses its virrun -- prefix once virrun retirement has landed, and every row below that mentions the prefix is gated on it.

The collapses

Four families account for most of the reduction, and in each case the script variants are encoding an argument.

The build family does not collapse as far as the others, and the exception is worth stating because it looks like an inconsistency. The bare build stays the app's, not the workspace's, for the deploy-entrypoint reason monorepo tooling gives; that reasoning is unchanged by the migration, so one specifically-named task survives for it.

What vp check absorbs

vp check runs formatting, linting and type-checking in one pass. That is three of the four steps this repository's own finishing ritual currently names separately, and the ritual is written down in more than one place — the agent guide's step ordering, the context-efficiency skill's rule about batching them at the end, and the release chain.

The release chain is the one that needs care rather than a find-and-replace. Every step in it is a check rather than a fix, for the reason monorepo tooling records under pnpm release. So the chain may adopt vp check only in its non-writing form, and the property to preserve is that a release fails on a dirty tree rather than tidying it. Any replacement that runs a formatter in fix mode is a regression, however much shorter it reads.

The ordering constraint inside that chain also survives: the format check is the only step that can run cold, because everything after it reads what the build writes — the generated barrels and dist are both gitignored, so from a clean checkout lint and typecheck resolve nothing and the size snapshots measure no bundle.

What vp subsumes outright

update:node is the interesting failure. It is not merely "install a Node version": it writes the two node pins and the matching type-definitions catalog entry together — which two, and why neither may be edited alone, is monorepo tooling — and the only other writer of them is Renovate's node group, which writes the same three.

vp env owns the runtime half. It does not know about this repository's second pin or its catalog entry, so either the script survives in reduced form — writing the pins and delegating the install — or the pins are proven redundant, which is a separate investigation with its own answer. Assuming vp env covers it is how one of the two pins silently goes stale, and the failure surfaces as CI provisioning the wrong runtime.

What does not move

• The sweep scripts. Each one is a find recipe for a convention sweep, with its own tests. They are repository tooling, not toolchain, and a task runner has nothing to say about them beyond running them.
• graph:gen. Generates the dependency graph image; bespoke and staying.
• release. Publishing stays with Lerna Lite, which is retained for publishing only and for nothing else. vp pack builds a library artifact; it does not do fixed-mode conventional-commit versioning across the public packages.
• bench, coverage, start, watch:packages. Thin wrappers over tools that Vite+ either already is or does not touch.

Windows-only scripts

crossOS and the PowerShell wrapper exist because development happens on a Windows host. If the virrun retirement decision moves that loop onto Linux, both are dead and should be deleted in the same change rather than left as a fallback — a fallback nobody exercises is the thing that breaks silently, and the repository's own standard is that superseded content is deleted rather than kept.

refresh:lockfile calls through crossOS and therefore inherits that decision. It is a real need either way, so it survives as a task; only its implementation stops needing a platform branch.

The documentation this changes

Every command spelling above appears in prose somewhere, and prose is what nothing checks — a stale path fails the Key Files test, a stale command name fails nothing. So the sweep is part of each phase and not a cleanup afterwards: the agent guide's command list, the package-scripts skill that owns script conventions, the context-efficiency skill's batching rule, and every area page that names a check by its script name. The full accounting is in docs cleanup.

Key files

Sources

• Vite+ and its monorepo guide — vp check, vp lint and vp run with a filter, the commands the script families collapse into.