\r \r

IFQ App Builder · OpenClaw / ClawHub Edition\r

\r Three sentences in → a verified, build-ready prompt bundle out — for PC desktop, macOS, Windows, iOS, Android, HarmonyOS, WeChat MiniProgram, self-hosted local web, PWA, or CLI. This ClawHub edition is a narrow, auditable skill package for OpenClaw: zero dependencies, no required credentials, no install hooks, workspace-scoped tools only.\r \r

Made by ifq.ai. Every generated bundle ends with the quiet colophon — shaped with ifq.ai/app-builder · \x3Cmode> · \x3Ctemplate>.\r \r

30-Second Load Path\r

\r

1. Confirm the user wants a real app or app-building prompt bundle. If the request is visual-only, route to a design skill instead.\r
2. Capture the three-sentence intent from references/three-sentence-contract.md: S1 WHO+WHAT, S2 WHERE, S3 SUCCESS.\r
3. If the user gave only one sentence, infer S2/S3 with (assumed) labels. Do not block with more than one clarifying question.\r
4. Pick the mode from references/modes.md, then read assets/templates/INDEX.json.\r
5. Fork the matched *.prompt.md template into the user's workspace and fill acceptance, packaging, security, i18n, and IFQ ambient sections.\r
6. Verify with npm run verify:lite -- \x3Cbundle.prompt.md> when shell is available.\r
7. Report the bundle path, mode, template, verification result, assumptions, and caveats that affect use.\r \r

First-Run Success Path\r

\r After install, the first user prompt should produce a visible *.prompt.md file, not a setup conversation:\r \r

1. Accept one natural-language app request, such as Create a Mac + Windows desktop tool that turns PDF invoices into an Excel ledger and runs locally with no network.\r
2. Route to exactly one mode and one template; name both in the evidence packet.\r
3. Write \x3Cworkspace>/\x3Cslug>.prompt.md with labeled assumptions where facts are missing.\r
4. Run npm run verify:lite -- \x3Cbundle.prompt.md> if OpenClaw shell permission is available.\r
5. Return: output file path, mode route, template id, assumptions made, verification command/result, and known caveats.\r \r No account login, global install, SDK installation, background daemon, or store submission belongs in the first-run path.\r \r

Use When\r

\r

• The user asks to build, make, ship, scaffold, or package a real app.\r
• The platform is desktop, macOS, Windows, iOS, Android, HarmonyOS, WeChat MiniProgram, local/self-hosted web, PWA, or CLI.\r
• The user wants a prompt bundle that another coding agent can execute end-to-end.\r
• The user says "three-sentence app", "cross-platform app", "fastest stable app build", or similar.\r \r

Do Not Use When\r

\r

• The deliverable is visual-only (slides, poster, infographic, brand board, landing mockup).\r
• The task is SEO marketing, copy editing, a backend service with no user-facing app, or a CSS-only bug fix.\r
• The user wants SDKs silently installed, certificates generated, or app-store submission claimed without evidence.\r \r

OpenClaw And ClawHub Contract\r

\r

• Install path: openclaw skills install ifq-app-builder.\r
• Local validation: npm run validate.\r
• Clean package: npm run pack creates a deterministic tarball outside the skill folder.\r
• Required plugins: filesystem, shell.\r
• Optional plugins: browser, memory.\r
• Filesystem: workspace-only read/write.\r
• Shell: bundled Node scripts only (validate, verify:lite, validate:templates, quality:score, pack).\r
• Network: optional outbound HTTPS fact checks only. No script-side network calls.\r
• Credentials: none required. Do not ask for signing certificates or store tokens in Tier 0.\r \r

Output Boundary\r

\r The ClawHub-safe output is a verified *.prompt.md build bundle. It does not run Xcode, Android Studio, Flutter, Rust/Tauri, DevEco, Docker, miniprogram upload tools, or store submission commands unless the user explicitly moves to implementation outside the Tier 0 skill loop.\r \r Never claim that an app, package, screenshot, store listing, or certificate exists unless the corresponding command returned 0 and the artifact exists on disk.\r \r

Tier Policy\r

\r | Tier | Default | Requires | Use for |\r |---|---:|---|---|\r | Tier 0 | yes | Node >= 18.17 | Generate, verify, score, and pack prompt bundles. |\r | Tier 1 | opt-in | Platform SDK chosen by caller | Actually scaffold and build the app described by the bundle. |\r | Tier 2 | opt-in | CI runners, certificates, store accounts | Notarization, signing, app-store submission, enterprise deployment. |\r \r

Reference Map\r

\r | Need | Load |\r |---|---|\r | Mode routing | references/modes.md, references/platform-matrix.md |\r | Three-sentence contract | references/three-sentence-contract.md |\r | Quality score | references/quality-bar.md |\r | Verification and exit codes | references/verification.md |\r | Packaging by platform | references/packaging.md |\r | i18n defaults | references/i18n.md |\r | Security baseline | references/security-baseline.md |\r | IFQ ambient layer | references/ifq-brand-spec.md |\r | Agent compatibility | references/agent-compatibility.md |\r \r

Safety Contract\r

\r

• No install-time execution hooks.\r
• No required environment variables.\r
• No dynamic code execution in skill scripts.\r
• No script-side network primitives.\r
• No process-spawning primitives in the ClawHub edition.\r
• No writes outside the active workspace.\r
• No secrets, .env, local OpenClaw state, VCS metadata, or binary assets in the published bundle.\r
• clawhub.json and this frontmatter intentionally duplicate OpenClaw metadata so loaders and humans can audit the same contract.\r \r

Delivery Evidence\r

\r When the skill finishes a user-facing run, report only evidence that exists:\r \r

• bundle file path\r
• mode route (A-xx)\r
• template id (T-...)\r
• assumptions made\r
• verification command/result\r
• caveats that affect use\r \r — shaped with ifq.ai/app-builder · OpenClaw · ClawHub\r