False positive being reported

[pronitdas]

I ran this on couple of vue 3 + vite projects. Mostly on the dead / unused-code checks

[vinayakkulkarni]

Wait, I still have to fix - #60 ; Thanks for the report though :)

[vinayakkulkarni]

Thanks for the detailed report — the screenshots made the root cause obvious. You're right: the two largest finding categories (dead-code/unused-file, dead-code/unused-export) were systematically wrong on projects using build-time auto-import/file-routing plugins.

Root cause

Doctor's dead-code pass uses knip. For plain Vue 3 projects it was only treating src/main.ts / src/App.vue as entry points, so every file resolved by a build-time plugin (no explicit import) looked unused:

• pages → unplugin-vue-router / file-based routing
• layouts → vite-plugin-vue-layouts
• components → unplugin-vue-components
• composables → unplugin-auto-import

Fix (shipped in 663aebe, refs 2026053/ce87189)

When doctor detects convention resolution (unplugin-auto-import, unplugin-vue-components, or vue-router present), it now registers src/pages, src/layouts, src/components, src/composables, src/stores as knip entry points, so those files are reachable and no longer flagged.

A subtlety worth recording: knip's programmatic createOptions() silently ignores entry/project passed as options (its Options type has no such fields, verified against knip 6.15.0 source). The config must be injected via a synthesized knip.json referenced through args.config — which is what doctor now does.

Verified before/after (real knip, on a fixture mirroring your setup)

	before	after
pages/, layouts/, components/, composables/ (auto-resolved)	flagged unused ❌	not flagged ✅
genuinely orphaned file (no import, no auto-resolve)	flagged ✅	still flagged ✅

So real dead code is still caught — only the auto-import/file-routing false positives are suppressed.

Still on the radar

• unused-export / unused-type-export for auto-imported composables/components can still over-report in some setups; tracking further refinement.
• unused-dependency false positives you flagged (Phase 1 · Rule: ai-slop/no-destructure-props-without-toRefs #10 in your table) are a separate knip dependency-resolution gap — investigating.

Leaving this open while I work through the remaining unused-* edges. Appreciate the report 🙏

[vinayakkulkarni]

Verified the export-level edge case in 6ff939e: added a regression guard asserting that the auto-imported useAuto export from composables/useAuto.ts is not flagged as dead-code/unused-export. The test passes against the existing fix — knip's entry-point semantics already cover the export case for auto-imported files (no source-level import is needed because build-time auto-import resolves the symbol).

Test Files  141 passed (141)
Tests       1527 passed (1527)
Coverage    100% (Stmts 3316/3316 | Branches 2307/2307 | Funcs 514/514 | Lines 2847/2847)
Lint        0/0

Still on the radar (per your earlier comment, not closed by this test):

• unused-type-export for auto-imported types — would need a fixture with an exported type alias + a separate assertion. Not blocking common cases; the symbol-resolution gap is in knip, not our config.
• unused-dependency false positives — separate knip dependency-resolution gap. Investigating on its own when prioritized.

Leaving this open per your note. Pinning the export edge case so a future buildKnipConfig change can't silently regress it.

[vinayakkulkarni]

Also pinned the type-export half of the same regression in a5da2f6: extended the fixture with an AutoOptions type export from composables/useAuto.ts + added a regression guard that knip must not flag it as dead-code/unused-type-export. The test passes against the existing fix (knip's entry-point semantics cover both function and type exports from auto-imported convention dirs).

Test Files  141 passed (141)
Tests       1528 passed (1528)  (+1 vs the function-export guard)
Coverage    100% (Stmts 3316/3316 | Branches 2307/2307 | Funcs 514/514 | Lines 2847/2847)

So both export-level edges of the #85 fix (function + type) are now regression-guarded. The unused-type-export concern in the earlier comment is no longer relevant — the fixture proves it's already handled.

One remaining edge from the earlier note, still separate work:

• unused-dependency false positives — separate knip dependency-resolution gap, not addressed here.

Leaving the issue open per your earlier note; unused-dependency is the only remaining category that may need a follow-up if it re-surfaces in user reports.