shadscan-vue

Find the UI fundamentals your shadcn-vue app forgot.

Deterministic checks. Evidence. Agent-ready fixes. Static audit — no AI required.

shadscan-vue.geoql.in — rule catalog, docs, and changelog.

shadscan-vue audits shadcn-vue (Vue 3 + Vite) and shadcn-nuxt (Nuxt 4) applications for the fundamentals that are easy to skip and expensive to miss: theme wiring, command menus, route and error states, accessible controls, form feedback, metadata, and mobile behavior.

npx shadscan-vue

It never starts your app, never edits files, never calls a model, and never uploads your source. Every finding cites a file and a line.

What you get

shadscan-vue v0.1.0
my-app · nuxt · pnpm

Score 66/100 · Grade D

Foundation             ████████████████░░░░ 84
Interaction            ██████████░░░░░░░░░░ 52
States                 ████████████████░░░░ 84
Accessibility          ████████████░░░░░░░░ 61
Forms and Data Entry   ███░░░░░░░░░░░░░░░░░ 18
Production Polish      ████████████████░░░░ 78

Findings

  FAIL images-have-alt (accessibility, error)
    <img> is missing alternative text.
      app/pages/index.vue:23
      fix: Add an alt attribute describing the image, or bind :alt to dynamic content.

Scoring

62 rules across six weighted categories:

| Category | Weight | | -------------------- | -----: | | Foundation | 20 | | Interaction | 20 | | States | 20 | | Accessibility | 20 | | Forms and Data Entry | 10 | | Production Polish | 10 |

Findings fall into four buckets. Fixes are provable failures that cost points. Advisories are lower-confidence observations that never subtract points. Passes are verified. Not applicable rules leave the score untouched — a project without forms is not penalised for form rules.

Grades: A ≥ 90, B ≥ 80, C ≥ 70, D ≥ 60, otherwise F.

Usage

shadscan-vue [path]                      # audit a project (defaults to .)
shadscan-vue --json                      # machine-readable report
shadscan-vue --prompt                    # paste-ready prompt for an AI agent
shadscan-vue --category accessibility    # run one category
shadscan-vue --fail-under 70             # exit 1 below a threshold
shadscan-vue rules                       # print the rule catalog
shadscan-vue setup --pre-commit          # install a git pre-commit hook

Exit codes are 0 (completed) and 1 (threshold not met, or an error).

Continuous integration

- run: npx shadscan-vue --fail-under 70 --no-interactive

--fail-under also fails when the score is unassessed or when source coverage was partial, so a silently-skipped scan cannot pass a gate.

Illustrative code is skipped, because a deliberately minimal example documents an API rather than shipping a surface: demo/, demos/, example/, examples/, __registry__/, and underscore-prefixed directories, alongside tests, fixtures, stories, and generated files. Matching is on directory names, so a component named DemoBanner.vue is still audited.

There is also a composite action, which writes the score and every failing rule to the job summary:

- uses: vinayakkulkarni/shadscan-vue@v0
  with:
    fail-under: 70

It takes path, version, fail-under, category, and summary, and exposes score, grade, and findings-count as outputs. See the repository readme.

Agent handoff

--prompt emits a remediation brief with the machine-readable report embedded in a delimited block, marked as untrusted data rather than instructions.

Supported projects

| Adapter | Detected by | | ------------- | ----------------------------- | | nuxt | a nuxt dependency | | vite-vue | vue and vite dependencies | | generic-vue | a vue dependency |

shadcn components are recognised through components.json aliases, through the components/ui convention, and through shadcn-nuxt auto-imports.

Credits

shadscan-vue is a Vue and Nuxt port of shadscan by TheOrcDev, built with the original author's permission. The rule taxonomy, scoring model, and product concept originate there. See THIRD_PARTY_NOTICES.md.

License

MIT © Vinayak Kulkarni. Portions derived from shadscan, MIT © 2026 TheOrcDev.