SaaS Dashboard Planner
Turn 31 real SaaS dashboards into an executable CLAUDE.md — every checklist item cited, no architecture guesswork.
The problem this kills
Every SaaS builder faces the same decision loop: which auth method to include, how to structure multi-tenancy, when to add webhooks, what billing shape to use. Without data, you guess. You build SAML SSO before your first enterprise customer. You add usage-based metering but forget to wire the event log from day one. You ship a four-tier role system when 24 of 31 production dashboards use exactly three tiers.
The result is a doom-loop: two weeks of architecture debates, half-built GDPR pages, a billing webhook that breaks on plan downgrades. Your coding agent keeps building because you keep describing from memory, and memory is vague.
Why your agent cannot fake this
The SaaS Dashboard Planner encodes what 31 real SaaS dashboards actually do. Not what the documentation says they do. Not what a language model predicts they might do. What they shipped.
The research window covered 9 open-source starters (Next.js, Supabase, Laravel starters including mickasmt, leerob, makerkit, supastarter), 22 live in-market products (Linear, Vercel, Stripe, Supabase, Cal.com, PostHog, Resend, Clerk, Neon, Inngest, Tailscale, Notion, Crisp, Cursor, Lovable, Webflow, Framer, Plain, BetterStack, Tinybird, Mintlify, Plausible), and a structured scan of top-selling CodeCanyon multi-tenant templates.
Every checklist item in the planner carries the N/31 count, the frequency percentage, the observed variant patterns across those 31 sources, and a curated decision rule. When the planner tells your agent "include 3-tier RBAC," it cites 24/31 dashboards. When it says "defer SAML SSO unless enterprise," it explains that every one of the 14/31 products that ships SAML gates it behind a Business or Enterprise tier.
Your own Claude cannot fabricate "found in 13/31 reviewed dashboards; wire event tracking from day one because retrofitting metering is painful" without doing the same scrape. That is the moat.
What you actually get
A structured CLAUDE.md planner that walks your coding agent through seven phases before writing a line of code:
- Forced questions that set every downstream default: buyer tier, pricing model, multi-tenant boundary, tech stack, trust gradient, launch shape
- Cited suggestion checklists for authentication, multi-tenancy, member permissions, billing, admin tooling, dashboard surface, settings tabs, notifications, developer surface, integrations, compliance, and operations stack
- Open prompts that capture your north star feature, the manual process being replaced, and your most frequent user action
- Synthesis rules that apply buyer-tier biases, pricing-model biases, and compliance biases to produce a complete CLAUDE.md
- A fill-in-the-blanks output template with 18 sections: purpose, product anatomy, tech stack, data model, page list with priorities, auth strategy, billing implementation, member and permission model, admin tooling, dashboard composition, notifications, developer surface, onboarding flow, operations stack, north star, manual process replaced, constraints and red lines, and implementation phases
The included example output shows a complete CLAUDE.md produced for a real buyer scenario, so you can see exactly what your agent will receive.
Coverage
Phase 1 - Forced questions: Seven structured questions that set the defaults for every subsequent decision. Buyer tier is the single most powerful lever: it shifts defaults across roles, compliance, SSO, and support tiers simultaneously.
Phase 2 - Cited checklists (14 categories): Authentication (11 options with N/31 cites), multi-tenancy (9 options), members and permissions (8 options), billing (15 options including the full Stripe webhook list), admin tooling (8 options), dashboard surface (9 options), settings tabs (derived default order from 28-90% frequency range), notifications (10 channels), developer surface (12 options including MCP server as emerging), integrations (pick-3 rule), compliance and trust, and operations stack.
Phase 3 - Open prompts: Five verbatim-capture prompts covering the differentiating feature, the manual process being replaced, the most frequent user action, architecture-shaping constraints, and stack red lines.
Phase 4 - Synthesis and output: Buyer-tier biases, pricing-model biases, compliance biases, and P3.4 architectural overrides applied. Output is a complete CLAUDE.md your agent drops into the repo and executes end-to-end.