Upgrade to TypeScript 7

[gingur]

Overview

Upgrade shared TypeScript toolchain configs from TypeScript 5 to TypeScript 7.

Current state

• typescript: ^5 in both dependencies and peerDependencies
• typescript-eslint: ^8
• No .ts source files — TypeScript is used as a shared config dependency for consumer repos

Tasks

• Update typescript dependency to ^7 (both regular and peer)
• Update typescript-eslint to a version compatible with TS 7
• Review and update any shared tsconfig base configs in configs/
• Verify ESLint/Prettier configs still work with the new version
• Coordinate with consumer repos (hooks, troyrhinehart) that depend on these shared configs

---

Tracked cross-repo task

• Migrate to oxlint/oxfmt + TypeScript 6 (lockstep with devkit Vite+ toolchain) troyrhinehart#21 — Migrate to oxlint/oxfmt + TypeScript 6 (not linkable as a sub-issue: bot has read-only access on gingur/troyrhinehart)

[github-actions]

⚠️ @gingur Planning turn did not complete (status: failure). See the run log: https://github.com/gingur/devkit/actions/runs/29109757387

Fix the cause, then reply here (any reply starts a new turn) or re-assign the bot — turns are safe to repeat.

[gingur-bot]

Plan

Ask (restated): Upgrade devkit's shared TypeScript toolchain (deps, peer deps, configs/) from TypeScript 5 to TypeScript 7, keeping the ESLint/Prettier configs working and moving the consumer repos (hooks, troyrhinehart) in lockstep.

Current state: package.json pins typescript: ^5 and typescript-eslint: ^8 in both devDependencies and peerDependencies; configs/tsconfig.base.json sets 9 modern flags; configs/eslint.config.mjs uses tseslint.configs.recommended. Consumers: gingur/troyrhinehart consumes @gingur/devkit#main with typescript ^5 / typescript-eslint ^8 / astro ^5; gingur/hooks does not consume the shared configs (only its own typescript ^5.7.0 devDep).

Approach: ⚠️ TS 7 is not adoptable today — recommending TS 6.0 now, 7 deferred. Verified against npm and microsoft/typescript-go: typescript@7.0.2 (npm latest) is the native Go compiler and no longer ships the JS compiler API (lib/typescript.js is gone; the new RPC API is listed "not ready"). typescript-eslint — latest 8.63.0 and canary — peer-requires typescript >=4.8.4 <6.1.0, so installing TS 7 breaks the shared ESLint config for every consumer. TypeScript 6.0 is the official bridge release: identical language behavior to 7.0, JS-based, and supported by typescript-eslint ≥ 8.58.0 (first release admitting TS 6). Plan: bump devkit to typescript ^6 + typescript-eslint ^8.58, then lockstep bumps in the two consumer repos (cross-repo task issues, created in those repos and linked as sub-issues here). When typescript-eslint's peer range admits TS 7, a small follow-up ask finishes the jump — I'd not pre-create that task since it isn't implementable yet.

Rejected alternatives:

• Straight to typescript@^7, dropping typescript-eslint — loses TypeScript-aware linting fleet-wide for no language-behavior gain over 6.0.
• Widen peers to ^5 || ^6 — devkit is consumed via #main; the standing convention is lockstep migration, and the fleet is two repos.
• Wait for typescript-eslint TS 7 support — indefinite timeline; 6.0 delivers TS 7 semantics now.

Proposed tasks

1. Bump devkit TypeScript baseline to 6.0 (deps, peers, docs) — in gingur/devkit

• Depends on: nothing
• Body:

Task 1 body (verbatim)

Context

Devkit's shared TypeScript toolchain pins typescript: ^5 and typescript-eslint: ^8 in both devDependencies and peerDependencies (package.json). The goal is TypeScript 7, but TS 7 is blocked today: typescript@7 (npm latest, 7.0.2) is the native Go compiler and no longer ships the JS compiler API that typescript-eslint requires (microsoft/typescript-go README lists API as "not ready"), and typescript-eslint@8.63.0 (latest and canary) peer-requires typescript >=4.8.4 <6.1.0. TypeScript 6.0 is the bridge release: same language behavior as 7.0, JS-based, supported by typescript-eslint ≥ 8.58.0 (first release whose peer range admits 6.0). Adopt 6.0 now; TS 7 lands in a later ask once typescript-eslint supports it.

Changes

In package.json:

• devDependencies and peerDependencies: typescript ^5 → ^6; typescript-eslint ^8 → ^8.58 (the floor matters: 8.0–8.57 peer-reject TS 6, and consumers inherit this range).
• Refresh pnpm-lock.yaml (pnpm install).

In README.md (peer-dependency table / consumer install docs, ~lines 49–59):

• Add a short migration note: consumers must bump typescript to ^6 and typescript-eslint to ^8.58 in lockstep, since devkit is consumed via #main. Note TS 7 is deferred pending typescript-eslint support for the new compiler API.

Review configs/tsconfig.base.json: every flag (strict, noUncheckedIndexedAccess, noImplicitOverride, noFallthroughCasesInSwitch, forceConsistentCasingInFileNames, esModuleInterop, resolveJsonModule, skipLibCheck, isolatedModules) is expected to remain valid in TS 6 — verify rather than assume; remove/adjust any flag TS 6 rejects or reports deprecated.

Verification (evidence required in the PR)

1. pnpm install completes with no peer-dependency warnings.
2. tsconfig.base.json valid under TS 6: in a temp dir, create a tsconfig.json extending configs/tsconfig.base.json plus one trivial .ts file; pnpm exec tsc --noEmit exits 0 with no deprecation warnings; include tsc --showConfig output.
3. Shared ESLint flat config parses TS under the new pair: pnpm exec eslint --config configs/eslint.config.mjs <sample.ts> runs clean.

Acceptance criteria

• package.json shows typescript: ^6 and typescript-eslint: ^8.58 in BOTH devDependencies and peerDependencies; pnpm-lock.yaml updated.
• README migration note present.
• PR description cites the three verification outputs above.

2. Lockstep TypeScript 6 bump — in gingur/troyrhinehart

• Depends on: task 1 (devkit's bump must land on main first)
• Body:

Task 2 body (verbatim)

Context

gingur/devkit (consumed here as @gingur/devkit via github:gingur/devkit#main) has bumped its TypeScript toolchain baseline from TS 5 to TS 6.0: its peer deps are now typescript: ^6, typescript-eslint: ^8.58. This repo pins typescript: ^5 and typescript-eslint: ^8 in devDependencies and must move in lockstep (devkit convention — devkit is consumed via #main, so the peer range change is already live). TS 6.0 is the official bridge release toward TS 7: identical language behavior, still JS-based, so Astro and typescript-eslint tooling keep working.

Changes

• package.json devDependencies: typescript ^5 → ^6; typescript-eslint ^8 → ^8.58.
• Reinstall to refresh the lockfile and pick up current @gingur/devkit#main.

Verification (evidence required in the PR)

• Install completes with no unresolved peer-dependency warnings.
• Typecheck passes under TS 6 (astro check or the repo's typecheck script).
• Lint passes using the devkit shared ESLint config (the repo's lint script).
• Build succeeds (astro build / the repo's build script).

Acceptance criteria

• Both devDependencies bumped as above; lockfile updated; PR description cites typecheck + lint + build output.

3. TypeScript 6 bump for fleet consistency — in gingur/hooks

• Depends on: nothing (independent — hooks does not consume devkit's shared configs)
• Body:

Task 3 body (verbatim)

Context

The gingur fleet is moving its TypeScript baseline from 5 to 6.0 (the official bridge release toward TS 7 — identical language behavior, still JS-based). This repo does not consume @gingur/devkit's shared eslint/tsconfig configs; it has its own typescript: ^5.7.0 devDependency, so this bump is independent and done for fleet consistency.

Changes

• package.json devDependencies: typescript ^5.7.0 → ^6; refresh the lockfile.

Verification (evidence required in the PR)

• Install completes cleanly; the repo's typecheck, build, and test scripts all pass under TS 6.

Acceptance criteria

• Dependency bumped; lockfile updated; PR description cites passing typecheck/build/test output.

What you might be missing:

• Tasks 2 and 3 are cross-repo — I'd create them as claude-task issues in gingur/troyrhinehart and gingur/hooks and link them as sub-issues here. If those repos' implement workflows aren't wired for claude-task, they'll sit as regular tracked issues until picked up — say so if you'd rather I keep everything in devkit or you handle consumers yourself.
• The ask's title ("Upgrade to TypeScript 7") is intentionally not what this plan delivers — 6.0 is the honest maximum today. If you want raw typescript@7 anyway (accepting broken typescript-eslint), reply and I'll re-plan.
• Watch condition for the TS 7 follow-up: typescript-eslint's typescript peer range admitting 7.x (currently <6.1.0).

To approve: tick Approve on the action panel below (posted automatically after this comment).
To change: just reply — any comment starts a revise turn.

---

This turn: researched the TS 7 landscape (npm registry, typescript-go repo) and repo/consumer state; posted this plan. Note: the previous run failed before producing anything — nothing to recover. Needed from you: approve, or reply with corrections (especially on the cross-repo task question above).

[gingur-bot]

🎛️ Your move

Ticking the box is the only way to approve. Any reply is discussion — it starts a revise/answer turn automatically. Prefix a comment with [hold] to comment without starting a turn.

(panel closed)

[gingur]

Change of scope (operator): don't do a TS7-only upgrade — change gears to the Vite+ toolchain: TypeScript 7 + oxlint (replaces eslint) + oxfmt (replaces prettier).

Re-plan against that:

• configs/: the shared eslint + prettier configs get replaced by shared oxlint/oxfmt equivalents (keep the tsconfig work from the current plan). Map our current rule intent onto oxlint's rule set — call out anything that has no oxlint equivalent instead of silently dropping it.
• node.verify.yml / verify tooling: lint + format-check steps move to oxlint/oxfmt invocations.
• Consumer impact: devkit consumers inherit these configs — the plan needs a migration note per consumer surface (lint-staged / husky pre-commit included, if they reference eslint/prettier).
• Keep the TS7 portion of the existing plan; this is additive re-scoping, not a restart.

Research current oxlint/oxfmt maturity as part of the revise — if oxfmt is not yet production-ready, say so and propose what to adopt now vs. what waits, rather than forcing it.

[gingur-bot]

Changelog: re-scoped per your comment — Vite+ toolchain (TS7 + oxlint + oxfmt) replaces the TS-only plan; eslint/prettier configs retired; verify workflow gains a format step; consumer tasks rewritten.

Plan (v2)

Ask (restated): Move devkit's shared toolchain to the Vite+ stack: TypeScript 7, oxlint replacing ESLint, oxfmt replacing Prettier — shared configs in configs/, format/lint steps in node.verify.yml, and per-consumer migration.

Current state: configs/ ships eslint.config.mjs (js + tseslint recommended), prettier.config.mjs (printWidth 100, singleQuote, semi, trailingComma all), lint-staged.config.js (prettier+eslint), tsconfig.base.json. Peers: eslint ^9, @eslint/js ^9, prettier ^3, typescript ^5, typescript-eslint ^8. node.verify.yml runs lint/typecheck/test scripts — no format-check step exists today. Consumers: gingur/troyrhinehart (Astro, uses all shared configs); gingur/hooks is already on Vite+ (vite-plus/vp check/vp test, which bundles oxlint+oxfmt) — only its typescript ^5.7 devDep is in scope.

Approach & maturity verdict (researched today):

• oxlint 1.73.0 — stable since 2025-06, strict semver on CLI/config, weekly minors, built-in eslint/typescript/react plugin rule sets, @oxlint/migrate converts our flat config. Adopt.
• oxfmt 0.58.0 — pre-1.0 versioning, but oxc positions it as the recommended dedicated formatter: passes 100% of Prettier's JS/TS conformance tests, native JS/TS/JSON/CSS/TOML/GraphQL, and md/yaml/html handled by a bundled Prettier (no separate dep). Vite+ itself ships it (pinned exact). Adopt now — pin deliberately (devDeps exact, peer ^0.58; 0.x caret pins the minor) to control formatting churn from weekly releases. Not forcing anything: nothing waits.
• TypeScript 7 — the v1-plan blocker was typescript-eslint's peer cap; dropping eslint dissolves it. oxlint needs no TS install at all; typescript@7.0.2 ships a working native tsc. One hard cap remains: @astrojs/check peers typescript ^5 || ^6, so troyrhinehart stays on TS6; devkit peers ^6 || ^7; hooks attempts 7 with a documented 6 fallback.
• Deferred: oxlint type-aware linting (oxlint-tsgolint, tsgo-based) — explicitly not semver-stable yet; follow-up ask when it stabilizes.

Rejected alternatives:

• Biome — Vite+ direction was specified; oxc is the Vite+ stack.
• Keep prettier for md/yaml only — unnecessary; oxfmt bundles Prettier for those formats.
• Force TS7 on troyrhinehart — would break astro check (peer cap); TS6 there until @astrojs/check admits 7.

Rule-intent mapping (your "no silent drops" requirement): current intent = js.configs.recommended + tseslint.configs.recommended (both syntax-only; no type-aware rules in play today). Task 1 requires the PR to include a generated-then-reviewed mapping table — every current rule → its oxlint rule or an explicit "no equivalent + disposition" row. Also called out per consumer: .astro files are neither linted nor formatted today (no astro plugins configured) and won't be after — no regression, documented.

Proposed tasks

1. Replace shared ESLint/Prettier configs with oxlint/oxfmt; TS baseline ^6 || ^7 — in gingur/devkit

• Depends on: nothing
• Body:

Task 1 body (verbatim)

Context

Devkit ships shared toolchain configs consumed by fleet repos via @gingur/devkit (github:gingur/devkit#main). We are moving to the Vite+ stack: oxlint (1.73.0, stable) replaces ESLint, oxfmt (0.58.0) replaces Prettier, TypeScript baseline moves to ^6 || ^7. Facts verified 2026-07-10: oxfmt passes 100% of Prettier's JS/TS conformance tests; md/yaml/html are formatted by a Prettier bundled inside oxfmt (no separate prettier dep needed); oxfmt defaults are printWidth 100, semi true, trailingComma "all" (singleQuote defaults false). typescript@7.0.2 ships a native tsc bin; nothing in this repo needs the TS JS compiler API once typescript-eslint is gone. oxlint config supports extends (paths resolved relative to the extending file). npx @oxlint/migrate (v1.73.0) generates .oxlintrc.json from an ESLint flat config. Docs: https://oxc.rs/docs/guide/usage/linter/config.md , https://oxc.rs/docs/guide/usage/formatter/config.md , https://oxc.rs/docs/guide/usage/formatter/migrate-from-prettier.md

Changes

configs/:

• Delete eslint.config.mjs and prettier.config.mjs.
• Add configs/oxlintrc.base.json: start from npx @oxlint/migrate run against the old configs/eslint.config.mjs (intent: js.configs.recommended + tseslint.configs.recommended), then hand-review. Keep the ignore patterns (dist/**, node_modules/**, coverage/**, .astro/**).
• Add shared oxfmt config carrying the old Prettier intent: { "printWidth": 100, "singleQuote": true, "semi": true, "trailingComma": "all", "sortPackageJson": false }. (sortPackageJson is ON by default in oxfmt — disable to avoid surprise churn; judgment call, revisit later.) Sharing mechanism: prefer a configs/oxfmt.config.mjs default-exporting the options object so consumers' oxfmt.config.ts can re-export it (mirrors the old prettier pattern); verify that works — fallback is consumers running oxfmt -c node_modules/@gingur/devkit/configs/.oxfmtrc.json. Document whichever mechanism survives verification in README.
• configs/lint-staged.config.js: '*.{js,mjs,cjs,ts,tsx,jsx}' → ['oxfmt', 'oxlint --fix']; '*.{json,md,yml,yaml,css}' → ['oxfmt'].
• configs/tsconfig.base.json: verify all 9 flags under TS7 tsc (remove/adjust anything rejected or deprecated).

package.json:

• exports: remove ./eslint, ./prettier; add ./oxlint → ./configs/oxlintrc.base.json and ./oxfmt → the oxfmt config file. Keep ./tsconfig, ./lint-staged, ./infisical-scan.
• peerDependencies: remove eslint, @eslint/js, prettier, typescript-eslint; add oxlint: ^1, oxfmt: ^0.58; typescript → ^6 || ^7.
• devDependencies: mirror (drop the four; add oxlint ^1.73, oxfmt pinned 0.58.x, typescript ^7). Refresh pnpm-lock.yaml.

README.md (~lines 43–59 consumer-config section, line 59 install one-liner, pre-commit section):

• Update the exports/peer table and install command (pnpm add -D oxlint oxfmt typescript).
• Consumer examples: .oxlintrc.json with { "extends": ["./node_modules/@gingur/devkit/configs/oxlintrc.base.json"] }; oxfmt wiring per the verified mechanism; scripts "lint": "oxlint", "fmt": "oxfmt", "fmt:check": "oxfmt --check".
• Migration note: consumers move in lockstep (devkit consumed via #main); TS7 for consumers without a TS-JS-API dependency, TS6 where one remains (e.g. @astrojs/check peers ^5 || ^6).

Rule mapping deliverable (no silent drops)

The PR description MUST include a mapping table: every rule enabled by js.configs.recommended and tseslint.configs.recommended → the oxlint rule that covers it, or an explicit "no oxlint equivalent" row with disposition (accepted-drop rationale). Generate from @oxlint/migrate output + the oxlint rules reference; review by hand.

Verification (evidence required in the PR)

1. pnpm install clean (no peer warnings).
2. Sample .ts file: pnpm exec oxlint -c configs/oxlintrc.base.json <sample> runs and flags a seeded violation (e.g. unused var).
3. Prettier-parity spot check: format a representative sample (ts + json + md) with old prettier config vs oxfmt; diff for ts/json is empty (md may differ slightly — report the diff).
4. tsc --noEmit (TS7) passes in a temp project extending configs/tsconfig.base.json; include tsc --showConfig.
5. lint-staged config resolves: stage a misformatted file, run npx lint-staged, file gets formatted+linted.

Acceptance criteria

• Old eslint/prettier config files deleted; oxlint/oxfmt configs exported; peers/devDeps as specified; lockfile updated; README updated; mapping table in PR; all five verification outputs cited.

2. node.verify.yml: add opt-in format-check step — in gingur/devkit

• Depends on: task 1 (README coherence; no code dependency)
• Body:

Task 2 body (verbatim)

Context

gingur/devkit's reusable workflow .github/workflows/node.verify.yml runs three consumer scripts: lint, typecheck, test. The fleet is adopting oxfmt for formatting; verify should be able to enforce oxfmt --check. Devkit is consumed via @main, so any unconditional new step would instantly break consumers that lack the script — the step must be opt-in via input.

Changes

In .github/workflows/node.verify.yml:

• Add input format (camelCase single word per repo conventions): description "pnpm script to run for format check. Empty skips the step.", type string, default ''.
• Add a Format step between Lint and Typecheck: if: inputs.format != '', run: pnpm ${{ inputs.format }} (same pattern as the other steps, honoring working-directory).

In README.md: update the verify recipe to show format: fmt:check and note it's opt-in.

Verification (evidence required in the PR)

• Workflow syntax validated: act run of a throwaway caller (or actionlint) showing (a) step skipped when format omitted, (b) step runs pnpm fmt:check when set.

Acceptance criteria

• Input + guarded step added exactly as above; omitting the input produces byte-identical behavior to today; README recipe updated; validation evidence cited in PR.

3. Migrate to oxlint/oxfmt + TypeScript 6 — in gingur/troyrhinehart

• Depends on: tasks 1 and 2 (devkit changes must be on main)
• Body:

Task 3 body (verbatim)

Context

@gingur/devkit (consumed here via github:gingur/devkit#main) has replaced its shared ESLint/Prettier configs with oxlint/oxfmt and moved its TypeScript peer to ^6 || ^7. This repo must migrate in lockstep. TypeScript stays at ^6 here — not 7 — because @astrojs/check (the typecheck script is astro check) peer-requires typescript ^5 || ^6. Note: .astro files were not linted/formatted before (no astro eslint/prettier plugins were configured) and won't be after — no regression; state this in the PR.

Changes

package.json:

• Remove devDependencies: eslint, @eslint/js, typescript-eslint, prettier.
• Add devDependencies: oxlint (match devkit's peer, ^1), oxfmt (match devkit's peer, ^0.58).
• typescript ^5 → ^6.
• Scripts: "lint": "oxlint"; add "fmt": "oxfmt" and "fmt:check": "oxfmt --check".
• Reinstall to refresh lockfile and pick up current @gingur/devkit#main.

Config files:

• Delete eslint.config.mjs (and any prettier config file).
• Add .oxlintrc.json: { "extends": ["./node_modules/@gingur/devkit/configs/oxlintrc.base.json"] }.
• Wire the shared oxfmt config per devkit's README (re-export via oxfmt.config.ts or -c flag — follow what devkit documents).

CI + hooks:

• In the caller workflow that uses gingur/devkit/.github/workflows/node.verify.yml@main, pass format: fmt:check.
• Husky pre-commit runs npx lint-staged with devkit's shared lint-staged config (now oxfmt/oxlint) — verify resolution still works; if lint-staged config is wired locally, point it at @gingur/devkit/lint-staged.

Verification (evidence required in the PR)

• Install clean (no peer warnings).
• pnpm lint, pnpm fmt:check, pnpm typecheck (astro check under TS6), pnpm build all pass.
• Seed one misformatted file: pnpm fmt:check fails, pnpm fmt fixes it, check passes.
• Stage a change and confirm the pre-commit hook formats/lints it.

Acceptance criteria

• Dep swaps exactly as listed; TS at ^6 with a PR note explaining the @astrojs/check cap; config files swapped; verify caller passes format: fmt:check; all verification outputs cited in the PR.

4. TypeScript 7 bump (6 fallback) — in gingur/hooks

• Depends on: nothing (independent — hooks is already on Vite+ and does not consume devkit's shared configs)
• Body:

Task 4 body (verbatim)

Context

The gingur fleet is moving its TypeScript baseline forward (target 7, minimum 6). This repo already uses Vite+ (vite-plus — vp check / vp test), which bundles oxlint/oxfmt internally, so no lint/format config changes are needed here. Only the typescript: ^5.7.0 devDependency is in scope. Unknown to resolve empirically: whether vp check (and nitro build) tolerates typescript@7 (the native compiler package — it no longer ships the JS compiler API some tools require).

Changes

• package.json devDependencies: typescript ^5.7.0 → ^7; refresh lockfile.
• Run the full suite: pnpm check (vp check), pnpm test (vp test --run), pnpm build (nitro build).
• If any of these fail because a dependency needs the TS JS API (typical symptom: cannot find/require typescript lib internals), fall back to typescript: ^6 and name the blocking dependency in the PR.

Verification (evidence required in the PR)

• Install clean; vp check, vp test --run, and nitro build outputs all passing at the final chosen TS major.

Acceptance criteria

• typescript at ^7 (preferred) or ^6 (with the blocker named); lockfile updated; all three command outputs cited in the PR.

What you might be missing:

• oxfmt churn risk: weekly 0.x minors can legally change formatting (esp. the Prettier-backed md/yaml). Mitigation baked into task 1: exact devDep pin + ^0.58 peer (0.x caret pins the minor); bumps become deliberate PRs.
• ESLint ecosystem exit is one-way-ish: if you ever want .astro-aware linting or a niche eslint plugin, oxlint's JS-plugins API is still alpha. Nothing in the fleet uses such plugins today.
• Editor setup drift: VS Code/WebStorm will keep using installed ESLint/Prettier extensions until pointed at oxlint/oxfmt LSPs — local-machine concern, not in these tasks.

To approve: tick Approve on the action panel below (posted automatically after this comment — do not add checkboxes here).
To change: just reply — any comment starts a revise turn automatically.

---

This turn: re-planned for the Vite+ scope — researched oxlint/oxfmt maturity (oxc.rs docs + npm), confirmed the oxlint pivot unblocks TS7 for devkit (typescript-eslint's peer cap was the only blocker), found @astrojs/check caps troyrhinehart at TS6, and discovered hooks is already on Vite+ (so its task shrinks to a TS bump). Cross-repo tasks (3, 4) will be created in their consumer repos and linked as sub-issues here, same as the v1 plan proposed. Needed from you: approve, or reply with corrections.

[gingur-bot]

🎛️ Your move

Ticking the box is the only way to approve. Any reply is discussion — it starts a revise/answer turn automatically. Prefix a comment with [hold] to comment without starting a turn.

(panel closed)

[gingur]

[hold] Forward-compat context for this plan (operator direction, no new turn needed): consumer repos are about to become pnpm workspaces with multiple packages (per-product monorepos — e.g. docpipe with packages/audit, packages/comms; looksfine with multiple workers). The shared configs this ask produces must be workspace-ready from day one: root-level oxlint/oxfmt/tsconfig that packages extend, no single-package assumptions. Also weigh standardizing on the single vp (Vite+) binary vs three separate dev-deps — Vite+ went MIT in March 2026, and its vite run task caching would persist across CI runs on our self-hosted runners. Workspace-aware node.verify and per-package cf.worker.deploy are explicitly OUT of scope here (future asks) — just don't preclude them.