lov-app-generator

app-generator — Skill Publisher App 生成器

Use this skill to create or upgrade a Skill Publisher-grade app. Choose the app type from the brief instead of forcing desktop packaging: use web-only when the workflow is browser-native, and use Tauri when native desktop capabilities or desktop distribution are required. Common stacks are React + TypeScript + Vite, Next.js, or Tauri + React, with shadcn/ui, Skill Publisher brand assets, Warm Academic UI, CI/CD or deploy wiring, and lovinsp click-to-code support.

Default Integration Invariant

Run the lov-integrate-lovinsp skill for every app handled by this workflow. Treat Lovinsp as a default development capability, not an optional feature selected from the brief.

• Invoke lov-integrate-lovinsp after the frontend scaffold and build config exist.
• Re-run it for existing apps to update Lovinsp or migrate code-inspector.
• Preserve its idempotent behavior; repeated app-generator runs must stay safe.
• Require the project audit to pass dependency, configuration, migration, and plugin-order checks before completion.
• Verify the served Vite module contains lovinsp-component or [lovinsp v...].

Skip this invariant only when the requested deliverable has no browser-rendered UI and therefore falls outside the app-generation paths described below.

Triggers

Activate when

• The user asks to generate a new Skill Publisher app, web app, PWA, desktop app, or cross-platform app.
• The user has an existing frontend/Tauri project and wants it brought up to Skill Publisher app standards.
• The user mentions web-only, Vite, Next.js, PWA, Tauri, shadcn, React Query / TanStack Query, deploy, auto update, CI/CD, app logo, Skill Publisher logo, Warm Academic UI, or lovinsp as part of app setup.
• The project is a Skill Publisher, Lovpen, Lovcode, Lovmind, Lovshot, Lovsider, Lovsigil, or Lovtarot app.

Do not activate when

• The user asks only for a standalone logo, document, presentation, or static media asset without an application shell.
• The request is limited to diagnosing an existing app and does not include creating or standardizing its app architecture.
• The deliverable is a backend-only service, CLI, library, or Skill package with no browser-rendered application UI.

Workflow (MANDATORY)

Resolve SKILL_DIR from the installed skill context before running helpers. For manual execution, set it to the directory containing this SKILL.md.

You MUST follow these steps in order:

Step 1: Clarify the App Brief

Collect only the missing fields. Use conversation context first. Prefer AskUserQuestion for interactive choices; if that tool is unavailable, ask short direct questions and continue once the answer is clear.

Required fields:

If the user asks for a real implementation and enough information is present, make conservative assumptions and proceed.

Suggested options to collect interactively:

Step 2: Read Local Context

Before changing files, inspect the target project:

pwd
find .. -name AGENTS.md -print
find .. -name CLAUDE.md -print
ls
find . -maxdepth 2 -type f \( -name package.json -o -name vite.config.ts -o -name next.config.ts -o -name next.config.js -o -name tauri.conf.json -o -name tauri.conf.json5 -o -name Cargo.toml \) -print

Honor any project-level instructions. If the target lives under a symlinked workspace, follow that project's own AGENTS.md / CLAUDE.md.

Step 3: Run the Skill Publisher App Audit

Run the helper from the target project root:

python3 "$SKILL_DIR/scripts/audit_app_project.py" --root . --app-type auto --format markdown

Use the output as the implementation checklist. For new projects, decide the app type first and pass --app-type web or --app-type tauri; the audit will mostly report missing pieces, which is expected.

Step 4: Choose the Implementation Path

App Type Decision

Pick the smallest app type that genuinely fits the brief:

• Web-only app: default when the product is a browser workflow, SaaS/admin surface, content or media tool, public site with logged-in tools, or anything that can deploy cleanly to Vercel/Netlify/Cloudflare/GitHub Pages.
• PWA: use when the app is still web-first but benefits from installability, offline shell, push notifications, or mobile home-screen usage.
• Tauri desktop app: use when native desktop value is explicit: local file access beyond browser capabilities, tray/menu/global shortcuts, long-running background tasks, native OS integration, offline-first packaged use, or GitHub Releases distribution with auto update.
• Framework choice: prefer Vite React for app-like single-page workflows, Next.js for SEO/SSR/content routing/API routes, and static HTML only for very small one-off deliverables.

Do not add Tauri simply because this skill historically defaulted to Tauri. If the user says "只创建 web" or the requirements do not need native desktop capabilities, create a web app.

New Web App

For an app-like browser workflow, default to Vite + React + TypeScript:

pnpm create vite@latest <project-slug> -- --template react-ts
cd <project-slug>
pnpm add @tanstack/react-query lucide-react
pnpm add -D typescript

For SEO-heavy, public, content-routed, or SSR/API-route requirements, use Next.js instead and keep the same Skill Publisher layers:

pnpm create next-app@latest <project-slug> --ts --tailwind --eslint --app --src-dir
cd <project-slug>
pnpm add @tanstack/react-query lucide-react

Then apply the Skill Publisher layers in this order:

1. Project identity: package name, app title, README, and app-specific CLAUDE.md / AGENTS.md.
2. Configurable Academic UI: shadcn/ui, semantic tokens, typography, and layout.
3. Server state: TanStack Query provider and query/mutation helpers when the app has server state; avoid unnecessary TanStack Query for purely local static tools.
4. Brand assets: generate a target-specific app logo with lov-gen-logo, publish the chosen version into assets/ and public/, and generate favicons / PWA icons if needed.
5. Lovinsp: invoke lov-integrate-lovinsp and verify click-to-code integration.
6. CI/CD and deploy: typecheck, lint/build where available, plus the selected web deploy target or documented manual deploy path.
7. Verification: typecheck, build, dev server, and browser screenshot or interaction check where practical.

New Tauri App

Use this path only when the app type decision requires native desktop capabilities or desktop distribution. Default stack:

pnpm create vite@latest <project-slug> -- --template react-ts
cd <project-slug>
pnpm add @tauri-apps/api @tanstack/react-query lucide-react
pnpm add -D @tauri-apps/cli typescript
pnpm tauri init

Then apply the Skill Publisher layers in this order:

1. Project identity: package name, app title, bundle identifier, README, and app-specific CLAUDE.md.
2. Configurable Academic UI: shadcn/ui, semantic tokens, typography, and layout.
3. Server state: TanStack Query provider, query keys, and Tauri invoke wrappers.
4. Brand assets: generate a target-specific app logo with lov-gen-logo, publish the chosen version into assets/ and public/, prepare a macOS-safe padded icon source, then run the Tauri icon pipeline from that generated logo.
5. Lovinsp: invoke lov-integrate-lovinsp and verify click-to-code integration.
6. CI/CD: typecheck, lint/build where available, Tauri release workflow.
7. Auto update: Tauri updater plugin, signing keys/env placeholders, release endpoint wiring.
8. Verification: typecheck, build, and app launch where practical.

Upgrade Existing App

Do not rebuild the project from scratch. Patch the smallest surface needed:

1. Determine whether the existing app should remain web-only, become a PWA, or stay/become Tauri before applying the audit checklist.
2. Keep the existing package manager, router, folder layout, aliases, and style conventions unless they conflict with Skill Publisher requirements.
3. Add missing Skill Publisher layers from the audit; do not add Tauri to a web-only app unless the brief requires native desktop capabilities.
4. Run lov-integrate-lovinsp to install/update Lovinsp and migrate any supported code-inspector integration.
5. Preserve user code and unrelated changes.
6. Prefer incremental commits/checkpoints when the app is already substantial.

Step 5: Apply Brand and UI Standards

New apps must not use the canonical Skill Publisher logo as the app/product icon. After the project identity and README describe the target clearly, invoke the lov-gen-logo workflow from the new app root:

1. Generate assets/logo-drafts/v1-*.png and .svg based on what the app does, not a literal reading of its name.
2. Publish the chosen draft to assets/logo.png, assets/logo.svg, public/logo.png, and public/logo.svg.
3. For web-only apps, generate favicons and PWA icons from the target-specific logo as needed; do not run Tauri icon tooling.
4. For Tauri apps, before feeding the generated logo into the Tauri icon pipeline, ensure the icon source has transparent safe area. Do not use a 512x512 edge-to-edge filled icon as the macOS app icon source; it appears oversized in Dock, Launchpad, and Finder. Prefer roughly 40-56px transparent padding on a 512x512 canvas, or a content bounding box around 80-85% of the canvas.
5. Use that padded generated logo as the source for lov-install-tauri-logo and any favicon/tray-icon generation.

For upgrades, keep an existing product logo unless the user asks to refresh it; if the app has no logo, use lov-gen-logo before generating icons.

Resolve canonical assets without assuming a private machine layout:

1. Explicit paths supplied by the user.
2. SKILL_PROFILE_PATH and SKILL_DESIGN_GUIDE.
3. ${SKILL_PROFILE_PATH:-$HOME/.skill-publisher/skills/profile.json}.
4. Ask once when a required asset is still missing.

See references/user-config.md for the portable configuration contract.

Rules:

• Treat the canonical Skill Publisher logo as brand reference or fallback only, not as the default app icon for new apps.
• For Tauri/macOS icons, verify the generated app icon is visually aligned with normal macOS app icons. If ImageMagick is available, a quick sanity check is: magick src-tauri/icons/icon.png -alpha extract -trim -format '%wx%h%O\n' info:; for a 512x512 source, content around 400x400 to 440x440 with positive offsets is usually safer than 512x512+0+0.
• Use semantic Tailwind classes such as bg-background, text-foreground, bg-primary, border-border; do not hard-code brand hex values in UI components.
• Keep the UI operational and app-like. Do not create a marketing landing page when the user asked for an app.
• Use shadcn/ui controls, lucide icons, compact panels, predictable navigation, and no nested cards.
• First screen should be the real product workflow.

When shadcn/ui is needed, use the existing lov-install-shadcn-ui skill as the detailed reference. When TanStack Query is needed, use lov-install-tanstack-query. For Tauri app icons, use lov-install-tauri-logo; for new apps, run lov-gen-logo first and feed the generated logo into the relevant favicon/PWA/Tauri icon pipeline.

Step 6: Tauri App Baseline

Only for Tauri desktop apps, check these areas:

Step 7: CI/CD, Deploy, and Auto Update

Default GitHub Actions surface for web-only apps:

• check.yml: install, typecheck, lint/build if present.
• Deploy wiring for the selected target: Vercel, Netlify, Cloudflare Pages, GitHub Pages, self-hosted static output, or a clearly documented manual deploy path.
• Environment variable placeholders only when the app actually needs them.

Default GitHub Actions surface for Tauri apps:

• check.yml: install, typecheck, lint/build if present.
• release.yml: Tauri build for target platforms, draft or publish GitHub Release, attach artifacts.
• Tauri updater wiring: plugin dependency, updater config, signing key env placeholders, and documented release process.
• For Tauri v2, plugins.updater.pubkey is required at runtime. Do not leave it out even during early scaffolding: a missing pubkey causes the app to panic during updater plugin initialization. Use a clear placeholder such as PLACEHOLDER_REPLACE_WITH_TAURI_SIGNER_PUBLIC_KEY until the real public key is generated with pnpm tauri signer generate.

Do not invent secrets. Use placeholder names and document where the user must set them:

• TAURI_SIGNING_PRIVATE_KEY
• TAURI_SIGNING_PRIVATE_KEY_PASSWORD
• platform signing/notarization secrets as required by the target app

Step 8: Lovinsp

Invoke the existing lov-integrate-lovinsp workflow for every browser-rendered app. Do not replace that workflow with a handwritten dependency-only installation: it also performs version checks, idempotent configuration, and code-inspector migration.

Completion requires all of the following:

• lovinsp exists in project dependencies.
• The supported build configuration registers lovinspPlugin.
• Vite registers lovinspPlugin({ bundler: "vite" }) before the framework plugin.
• No supported legacy code-inspector dependency or configuration remains.
• A development-server readback proves the transform is active when practical.

For Vite apps, confirm the Vite config imports and registers lovinspPlugin({ bundler: "vite" }) before the framework plugin, not merely that the package is installed. In dev mode, verify the served module contains lovinsp-component or [lovinsp v...]:

curl -s http://127.0.0.1:<port>/src/main.tsx | rg "lovinsp-component|lovinsp v"

For web-only apps, start the dev server and provide the local URL when the user needs to try the app:

pnpm dev --host 127.0.0.1

For Tauri apps, prefer launching dev mode through a persistent session when the user wants to keep it running after the turn:

tmux new-session -d -s <slug>-dev -c "$PWD" 'pnpm tauri dev'
tmux capture-pane -pt <slug>-dev -S -120

Step 9: Verification

Run the lightest reliable checks that the target repo supports:

pnpm exec tsc --noEmit --pretty false
pnpm build
# Tauri only:
pnpm tauri build --debug

Adjust for npm/yarn/bun and local instructions. If a dev server is needed to verify frontend behavior, start it and give the user the local URL.

For UI changes, use browser or screenshot verification when practical. For Tauri-native behavior, report what was and was not verified.

User Configuration

If local paths differ, prefer these environment variables rather than hard-coding personal paths:

CLI Reference

Dependencies

python3 "$SKILL_DIR/scripts/audit_app_project.py" --help

No Python packages are required.

Final Response Checklist

Report:

• App path and stack chosen.
• App type decision: web-only / PWA / Tauri, and why that fit the brief.
• Skill Publisher layers added or confirmed: brand, UI, data layer, lovinsp, CI/CD, deploy/release, and updater only when applicable.
• lov-integrate-lovinsp result, including installation/update/migration status and runtime readback evidence.
• Commands/checks run and their result.
• Any remaining secrets, signing steps, or manual app-store/release actions.

Runtime context (shared)

运行前读取本 Skill 包的 skill.yaml，由宿主提供 skill-runtime/v1 上下文。字段解析顺序为：当前请求、项目上下文、个人 Preferences、品牌 Profile、通用默认值。

• 只使用 Manifest 声明的字段；Profile 保存公开品牌事实，Preferences 保存个人工作偏好。
• required: true 字段缺失时，按 Manifest 的问题配置向用户提出一个聚焦问题；用户明确同意后再保存回答。
• 报错提供可复制的 context_id、字段路径与来源，诊断内容避开秘密、完整私人路径和原始配置。

通用反馈闭环

用户在 Skill 驱动任务中提出修改意见时，继续当前产物前必须执行：

1. 先判断意见是 task-specific（仅本次）还是 reusable（可跨任务复用）。
2. task-specific 只修改当前任务，不改 Skill。
3. reusable 先确定作用域：领域规则先更新对应 canonical Skill；适用于所有 Skill 的规则先更新共享规范。
4. 完成规则更新、版本、lint 与分发核验后，再把修改应用到当前任务。
5. reusable 修改会使此前的“确认”“继续”“发吧”失效；完成当前产物修改和回读后必须停下，等待用户下一步指示，不自动进入发布、提交或其他外部写入。