chore(ui): make knip trustworthy and enforce dead-code in CI

[ryan-crabbe-berri]

Relevant issues

Linear ticket

Pre-Submission checklist

Please complete all items before asking a LiteLLM maintainer to review your PR

• I have added meaningful tests
• My PR passes all CI/CD checks (e.g., lint, format, unit tests)
• My PR's scope is as isolated as possible; it only solves 1 specific problem
• I have received a Greptile Confidence Score of at least 4/5 before requesting a maintainer review (Greptile reviews automatically once the PR is opened; only comment @greptileai to re-request a review after pushing changes)

Screenshots / Proof of Fix

knip was already in the repo but never run in CI, so it had quietly rotted. Running it exposed two failure modes that made it untrustworthy: it silently degrades on a partial install (a fresh worktree that had not run npm install flagged all 436 test files as unused), and even on a complete tree its config had blind spots that surfaced as false positives.

Before, at base 27cf064556 (full npm ci tree, previous config):

$ npm run knip
Unused files (2)
e2e_tests/serverRootPath.config.ts     <- FALSE POSITIVE; invoked via --config in test_server_root_path.yml
src/components/ui/select.tsx           <- FALSE POSITIVE; shadcn primitive, added to the palette ahead of use
Unlisted dependencies (198)
@ant-design/icons  ...                 <- all 198 are the same missing package.json entry
Unused exports (118)
Unused exported types (158)
...
$ echo $?
1

After, at 012afc0c4d (this branch):

$ npm run knip
Unused exports (100)
Unused exported types (158)
Unused exported enum members (6)
Duplicate exports (23)
$ echo $?
0

Both false-positive files and the 198 phantom "unlisted" entries are gone, and the exit code is 0 because the remaining findings are staged as non-blocking warnings. The CI job installs with npm ci, so it always runs against a complete tree and can never silently degrade

Type

🚄 Infrastructure

Changes

The goal is to make a green knip run actually mean something, then enforce it, rather than blindly deleting everything it lists (some of which is wrong)

Config correctness, so findings are real:

• Register every playwright config (serverRootPath.config.ts and migration.serverRootPath.config.ts), not just the main one. serverRootPath.config.ts is invoked via --config in test_server_root_path.yml, which knip cannot statically see, so it was wrongly reported as an unused file
• Treat src/components/ui/** as entry points. These are shadcn design-system primitives; some are intentionally added to the palette before every one has a consumer, and each exports its full sub-component family. knip was flagging the not-yet-used ones (e.g. select.tsx) as dead files and their exports as unused. Registering the directory as the design-system surface is the correct fix. The tradeoff is that knip will no longer flag a genuinely-abandoned ui/ primitive, which is an acceptable call for a vendored component kit
• Declare @ant-design/icons as a direct dependency. It is imported in ~198 files but only resolved through antd's hoisting, so each import showed up as an "unlisted dependency". This is a real fragility fix, not just noise suppression
• Add an explicit vitest plugin block so test-file classification no longer depends on auto-detection (the exact thing that broke on partial installs)

Governance, so it ratchets like the existing lint budgets:

• Stage severities via rules: gate the categories that are clean today (files, dependencies, unlisted, unresolved) as errors, and keep exports/types/duplicates as warnings so they are visible without blocking. Follow-ups can flip each category to error as it is driven to zero
• Run npm run knip in the frontend-lint CI job
• Move frontend-lint (prettier + eslint + budgets + knip) out of test-litellm-ui-build.yml and into its own test-litellm-ui-lint.yml named "UI Lint". It was sharing the build workflow's file, so its check surfaced as the misleading "UI Build Check / frontend-lint" despite doing no building; it now reads "UI Lint / frontend-lint". Whoever manages branch protection needs to update the required-check name from the old context to the new one so PRs don't strand

No code is deleted. Every file knip flagged turned out to be a false positive traceable to a config blind spot, which is the whole point: fix the tool so its output is trustworthy, rather than delete real code to satisfy a mis-tuned linter

The larger Unused exports (100) / Unused exported types (158) lists are left for follow-up PRs; they need per-item verification (knip's static analysis has the same blind spots that produced the serverRootPath false positive here), so blind-deleting them is exactly what this PR is trying to avoid

[greptile-apps]

Greptile Summary

This PR makes the existing knip dead-code scanner trustworthy by fixing config blind spots that caused false positives, then enforces it in CI so a green run actually means something. No production code is deleted.

• Config correctness: registers all three Playwright configs, treats src/components/ui/** as design-system entry points (so un-consumed shadcn primitives aren't mis-flagged), and adds an explicit vitest plugin block to prevent partial-install degradation.
• Explicit dependency declaration: promotes @ant-design/icons from an implicit transitive hoist to a direct package.json entry, eliminating the 198 phantom "unlisted dependency" warnings.
• Ratcheted enforcement: a new rules block in knip.json gates the now-clean categories (files, dependencies, unlisted, unresolved, etc.) as errors and keeps noisier categories (exports, types, duplicates) as warnings; the new CI step runs npm run knip inside the frontend-lint job against a full npm ci tree.

Confidence Score: 5/5

Safe to merge — changes are limited to CI configuration, a linter config file, and a package.json dependency declaration; no production source code is touched.

The diff is purely infrastructure: a CI step that runs knip, a knip config that removes documented false positives, and a package.json entry that formalises an already-used transitive dependency. The PR description includes before/after output confirming exit code 0, and the config choices are clearly reasoned. Nothing here affects runtime behaviour.

No files require special attention.

Important Files Changed

Filename	Overview
.github/workflows/test-litellm-ui-build.yml	Adds a knip dead-code check step to the frontend-lint job; correctly inherits the job-level working-directory: ui/litellm-dashboard and uses the same has_files guard as the existing lint steps.
ui/litellm-dashboard/knip.json	Fixes three false-positive sources (shadcn ui/ entry points, multiple playwright configs, explicit vitest plugin) and introduces a rules block that gates clean categories as errors and keeps noisy ones as warnings.
ui/litellm-dashboard/package.json	Promotes @ant-design/icons from an implicit transitive hoist to an explicit direct dependency at 5.6.1, eliminating 198 false "unlisted dependency" warnings from knip.
ui/litellm-dashboard/package-lock.json	Lockfile updated to record @ant-design/icons 5.6.1 as a direct dependency entry; no other substantive changes.

_{Reviews (2): Last reviewed commit: "chore(ui): make knip trustworthy and enf..." | Re-trigger Greptile}

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[ryan-crabbe-berri]

@greptileai re review

[codspeed-hq]

Merging this PR will not alter performance

✅ 31 untouched benchmarks

---

_{Comparing litellm_knip_confidence (47c3280) with litellm_internal_staging (bf02a4a)¹}

Footnotes

1. No successful run was found on litellm_internal_staging (190ea08) during the generation of this report, so bf02a4a was used instead as the comparison base. There might be some changes unrelated to this pull request in this report. ↩

[mateo-berri]

LGTM; thanks!