agentleFS
MartinOliveroMartinOlivero/saas-builder/skills/saas-router/SKILL.md

saas-router

Skill4 starsChanged 32 days agoMITView on GitHub

This skill should be used FIRST whenever the user wants to create, build, design, ship, or prototype any digital product — a SaaS, web app, platform, MVP, landing page, sales page, dashboard, admin panel, backoffice, internal tool, API, database, auth, payments, an integration, an AI feature, or any user interface. Trigger phrases (English) include "build me a", "I want to create", "design a", "let's make a", "I have an idea for", "help me ship". Trigger phrases (Spanish) include "quiero construir", "armar una", "diseñar una", "hagamos un", "tengo una idea para", "necesito una app/SaaS/landing", "agregá login/pagos/base de datos/IA", "integrá con", "subir a producción". It routes the request to the correct specialist skill instead of writing code or designing blindly.

The file

We split this skill into the rules it gives an agent

4 rules. About 0 are common, 3 are distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Do not interrogate. Ask only what is needed to classify the request.

    line 20
  2. Distinctive:

    If the user's first message already answers these (e.g.

    line 26
  3. Distinctive:

    When routing, pass along everything already learned so the specialist does not re-ask.

    line 100
  4. Very specific:

    If the user explicitly asks for one specific thing ("just give me a Tailwind navbar"), honor it and go straight to ui-design.

    line 120

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

SaaS Router

This skill is the brain of the saas-builder plugin. It runs before any product work and decides which specialist skill should handle the request.

The one rule that cannot be broken

Do NOT write code, design a UI, or pick a stack before routing through this skill. Jumping straight to implementation is the most common failure mode — it produces generic output that ignores what the user actually needs. Route first, build second.

Analogy: this is the triage nurse at a hospital. Nobody goes straight to surgery. The nurse asks two or three quick questions, then sends the patient to the right specialist.

Workflow

Step 1 — Ask 2-3 quick routing questions

Do not interrogate. Ask only what is needed to classify the request. Pick from:

  • "What are you building — a landing page, a dashboard, or a full app/SaaS?"
  • "Do you already know exactly what it should do, or are you still shaping the idea?"
  • "Is there an existing product/codebase, or are we starting from zero?"

If the user's first message already answers these (e.g. "build me a pricing landing page for my course"), skip straight to routing — do not re-ask what is already known.

Step 2 — Route to the right specialist

Match the request (English or Spanish) against this table and activate the matching skill. The plugin covers the whole build lifecycle — discovery, architecture, build, secure, monetize, polish, ship.

Plan & design

User says / wantsRoute to
"SaaS", "platform", "app", "MVP", "product" / "idea para", idea not fully definedproduct-discovery → architecture-primer → build skills → ui-design
"architecture", "how do I structure this", "scale", "what stack" / "cómo lo estructuro", "qué stack"architecture-primer
"landing", "sales page", "lead capture" / "página de ventas", "captar leads"landing-page (structure + copy) + design-taste-frontend (visual execution) — and the landing-moment plugin for the hero, if installed
"portfolio", "marketing site", "agency site", "editorial/blog", "make a website that doesn't look AI-generated", website redesign / "portfolio", "sitio de marketing", "que no parezca hecho por IA", "rediseñar mi web"design-taste-frontend
"dashboard", "admin", "backoffice", "internal tool", "data table", product/app UI / "panel", "tablero", "interfaz de la app"ui-design (dashboard mode)
Component, colors, design system for an app, dark mode for product UI / "componente", "design system de la app"ui-design
"copy this design", "make it look like this screenshot/site", reference image of a web they like / "cloná esta web", "hacelo como esta captura"image-to-code

Web vs. app boundary (important): design-taste-frontend owns public-facing sites — landings, portfolios, marketing, editorial, redesigns — and explicitly does NOT do dashboards, data tables, or multi-step product UI. The moment the work is logged-in product UI (dashboards, admin, settings, data-heavy screens), route to ui-design instead. For a landing page, run both: landing-page sets the section order and copy, then design-taste-frontend executes the visual so it doesn't look templated.

Build the backend

User says / wantsRoute to
"design the API", "endpoint", "REST or GraphQL" / "diseñar la API", "un endpoint"api-design
"database", "schema", "model this data", "multi-tenant" / "base de datos", "modelar los datos"data-modeling
"login", "auth", "roles", "permissions", "SSO" / "agregar login", "roles", "permisos"auth
"organizations", "workspaces", "teams", "invite members", "a user in two companies", "demo account" / "varias empresas en la misma app", "invitar usuarios", "cuenta demo"multi-tenancy → then data-modeling
"integrate with", "webhook", "sync with their API", "connect Google/Meta account", "transactional email" / "integrar con", "recibir un webhook", "sincronizar con", "mails transaccionales"integrations
"add AI to the app", "chatbot", "AI assistant for my users", "summarize with AI", "AI costs", "bring your own key" / "agregar IA", "armar un chatbot", "el bot inventa cosas"ai-features

Secure & monetize

User says / wantsRoute to
Writing any endpoint / input / data write, "is this secure", "rate limit", "secrets" / "esto es seguro", "manejar datos de usuarios"secure-coding (runs alongside backend work)
"payments", "Stripe", "Mercado Pago", "subscriptions", "pricing", "billing", "free trial" / "cobrar", "suscripciones", "precios"payments
"audit my RLS", "is my database exposed", "can users see each other's data", "anon key", "review my policies" / "auditá mi RLS", "¿se puede leer mi base con la anon key?"rls-audit
"how do I know this works", "is this actually tested", "prove the payment works" / "¿cómo compruebo que funciona?", "probalo de verdad"proof-first — separate plugin, see note below
"terms and conditions", "privacy policy", "GDPR compliance", "audit my legal docs", "do I need a DPO" / "términos y condiciones", "política de privacidad", "legales", "botón de arrepentimiento", "auditá mis legales"legal-docs — separate plugin, see note below

legal-docs vive en su propio plugin, no en saas-builder: /plugin marketplace add MartinOlivero/saas-legal-docs → /plugin install legal-docs

Está aparte porque la normativa envejece a otro ritmo que el resto de este plugin, y una tabla legal desactualizada hace daño real. Cubre Argentina, UE, EEUU, Brasil y Reino Unido, cada una con su tabla verificada contra fuente primaria y con fecha.

Si no está instalado, NO improvises los legales. Escribir una política de privacidad de memoria produce justo lo que esa skill existe para evitar: un documento que describe otro producto y cita normas que quizá ya no rigen. Decile al usuario que instale el plugin. Si insiste en avanzar igual, lo único responsable sin él es (a) pedirle razón social, identificación fiscal, domicilio y correo de contacto, que son obligatorios en toda jurisdicción, (b) no afirmar nada que no hayas verificado en el código, y (c) advertirle que el borrador no está respaldado por normativa verificada.

Companion plugins by the same author — optional, never block on them:

  • proof-first (/plugin marketplace add MartinOlivero/proof-first → /plugin install proof-first) — refuses to take the agent's word that money, auth, user data or a deploy works; sends a clean-context verifier for evidence observable from outside. If it is not installed, use the outside-in checks in pre-ship-security Step 2b.
  • landing-moment (/plugin marketplace add MartinOlivero/landing-moment → /plugin install landing-moment) — the method for a landing page's one authored interaction, with a cliche detector and a real-browser auditor. If it is not installed, landing-page carries the principle.

Polish & ship

User says / wantsRoute to
"slow", "performance", "Core Web Vitals", "bundle size" / "está lento", "optimizar"frontend-performance
"accessible", "a11y", "WCAG", "screen reader", "keyboard" / "accesibilidad"accessibility
"SEO", "meta tags", "social preview", "sitemap" / "que Google lo encuentre", "preview en WhatsApp"technical-seo
"PWA", "installable", "offline", "push notifications" / "que funcione offline", "instalable"pwa
"is this safe to ship", "pre-launch security check", "did I miss anything", before deploying / "antes de subir, ¿está seguro?"pre-ship-security
"deploy", "CI/CD", "GitHub Actions", "Sentry", "rollback", "staging" / "subir a producción", "monitoreo"deployment

Step 3 — Hand off with context

When routing, pass along everything already learned so the specialist does not re-ask. State the handoff explicitly, e.g. "This is a full SaaS with a fuzzy idea, so I'll start with product-discovery to lock the MVP, then architecture, then build."

The full-SaaS lifecycle

A from-scratch SaaS runs through these phases. Each feeds the next — don't skip, but don't force every phase if the user only asked for one slice.

  1. product-discovery — define the problem, the user, and the MVP scope.
  2. architecture-primer — decide the high-level system design and stack.
  3. Build — multi-tenancy first if customers are organizations (the membership model shapes everything after it) → data-modeling (the schema) → api-design (the endpoints) → auth (who can do what) → ui-design (the interface). Add integrations when the product talks to a platform you don't control, and ai-features when it calls a model for its users. secure-coding runs throughout, not as a final step.
  4. Monetize — payments when it's time to charge.
  5. Polish — frontend-performance, accessibility, technical-seo, pwa as the product matures.
  6. Legal — the legal-docs plugin (separate install) once the schema, the providers and the billing are settled: the documents are deduced from what the product actually stores and from where, so they can't be written earlier.
  7. Ship — pre-ship-security for a final security review of the finished code — plus rls-audit if the backend is Supabase, InsForge or another Postgres with an auto-generated API — then deployment for CI/CD, monitoring, and rollback.

Field notes. Most skills carry a references/field-notes.md: failures from production apps that passed build, lint and tests, each with symptom, cause and how to check for it. When a specialist skill is active and the work is about to ship, read its field notes.

Building UI before the MVP is defined, or picking a stack before knowing the scale, produces rework. But secure-coding is never deferred — preventing a vulnerability is always cheaper than auditing one out later.

When NOT to route

If the user explicitly asks for one specific thing ("just give me a Tailwind navbar"), honor it and go straight to ui-design. Routing serves the user — it is not a toll booth. Use judgment: ambiguity means ask; clarity means act.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.