Cheggin/startup-init

/startup-init — Autonomous Startup Builder

Take a startup idea from zero to running company. Research, design, build, deploy, grow. The user is CEO, not engineer. They get Slack investor updates, not questions.

Read .harness/stacks.yml for stack. Read .harness/agent-categories.yml for rules. Read .harness/tool-catalog.yml for integrations. All are law.

---

Execution Rules

• Phases run sequentially. Post Slack update after each phase.
• Track ALL work in GitHub Issues + Project board (Backlog | In Progress | In Review | Done).
• On context reset: gh issue list --state open --json number,title,body,comments to rebuild state.
• Coding phases: feature branches + PRs + Cubic review. Never push to main.
• Content phases: drafts first, check against SOUL.md.
• Never use Inter font, sparkles icon, !important, or left outlines.
• Constants in dedicated files. Reusable styles in global CSS. Fuzzy search for search features.
• Never run yarn dev or npm run dev. Use npm (not yarn, not pnpm).
• Auth is deferred until product spec confirms it. TDD always.
• Questions asked ONE AT A TIME. Never batch.

---

PHASE 0 — Deep Founder Interview

This is not "what's your idea?" It is a comprehensive diagnostic. Ask each question one at a time. Wait for the answer before asking the next. Adapt follow-ups based on prior answers. Skip questions already answered.

Interview Questions (sequential, one at a time)

Q1: The Idea "What's your startup idea? One sentence to one paragraph." Save raw answer to .harness/idea.md.

Q2: Company Type "What type of company is this?" Options: B2C consumer app | B2B SaaS | DevTool (CLI/SDK/API) | Infrastructure | Marketplace (two-sided) | Hardware + Software | E-commerce | Fintech | Healthcare/Biotech | AI/ML product | Content/Media platform | Other Store as startup_type in .harness/config.yml.

Q3: Target Users "Who are the target users? Be specific — a job title at a company type, not a category." Push if vague. "Developers" is not specific. "Frontend engineers at mid-size SaaS companies frustrated with CSS-in-JS performance" is specific.

Q4: Business Model "How does this make money?" Options: Subscription (monthly/annual) | Usage-based (pay per API call, etc.) | Freemium (free tier + paid) | Marketplace commission | One-time purchase | Ad-supported | Open-core (open source + paid features) | Enterprise licensing | Not sure yet Store as business_model.

Q5: Budget & Funding "What's your budget situation?" Options: Bootstrapped (minimal spend) | Pre-seed/angels ($0-$500K) | Seed ($500K-$2M) | Series A+ ($2M+) | Enterprise-backed (internal project) Determines: infra choices, monitoring tier, tool selection.

Q6: Technical Level "What's your technical background?" Options: Non-technical (no coding) | Technical but not coding daily | Full-stack developer | Specialized (backend only, frontend only, ML/AI, etc.) Determines: how much to explain, how autonomous to be, Slack update detail level.

Q7: Existing Tools "What tools and services do you already use?" Probe for: GitHub account? Vercel? Domain registrar? Figma? Slack workspace? Stripe? AWS/GCP? Any existing codebase? Pre-populates Phase 1 service connections.

Q8: Compliance Requirements "Any regulatory or compliance requirements?" Probe based on startup_type:

• Healthcare → HIPAA, PHI handling, BAAs
• Fintech → PCI DSS, KYC/AML, state licensing
• Enterprise SaaS → SOC 2, GDPR, data residency
• E-commerce → PCI (if handling cards), sales tax
• Children/education → COPPA
• None identified → confirm "no special compliance needed" Store as compliance[].

Q9: Hardware Component Only ask if startup_type includes hardware or if Q1 answer mentions physical products. "Tell me about the hardware. What does the device do? What connectivity? What sensors/actuators?" Probe: BOM estimate? Manufacturing partner? Firmware platform (ESP32, nRF, STM32, Raspberry Pi)?

Q10: Integrations Needed "What third-party services need to connect?" Probe based on startup_type:

• Payments → Stripe (default), alternatives?
• Maps/location → Google Maps, Mapbox?
• Messaging → Twilio, SendGrid, Resend?
• Search → Algolia, Typesense?
• AI/ML → OpenAI, Anthropic, Replicate?
• CRM → HubSpot, Salesforce?
• Calendar → Google Calendar, Cal.com? Cross-reference answers against .harness/tool-catalog.yml.

Q11: Timeline "What's your timeline?" Options: Weekend hack (2-3 days) | Sprint (1-2 weeks) | Month | Quarter | Ongoing (no deadline) Determines: P0 scope aggressiveness, whether to skip P1/P2 features initially.

Q12: Existing Designs "Do you have any designs already?" Probe: Figma file URL? Screenshots? Wireframes? Napkin sketches? Inspiration sites? If Figma URL: use mcp__figma__get_design_context to extract immediately. If screenshots: save to .harness/design-inputs/. If inspiration: save URLs to .harness/config.yml under design.inspirations[].

Q13: Domain Name "Do you have a domain name? Or a name for the startup?" If yes: validate availability via web search. Store in config. If no: note "domain TBD" — will be resolved in Phase 4.

Interview Output

Write .harness/founder-profile.yml:

idea: <raw idea>
startup_type: <type>
target_users: <specific description>
business_model: <model>
budget: <tier>
technical_level: <level>
existing_tools: [<list>]
compliance: [<list>]
hardware: <null or description>
integrations: [<list>]
timeline: <timeline>
designs: <null, figma_url, or path>
domain: <name or TBD>
interview_date: <date>

Slack: "Founder interview complete. Building: {idea}. Type: {type}. Timeline: {timeline}. Starting service connections."

---

PHASE 1 — Service Connections

Walk through each service. Validate before proceeding. Skip services the founder already confirmed in Q7.

Agent: Commander (orchestrates)

1.1 GitHub

Validate: gh auth status
Extract: username, org
Fail: "Run `gh auth login` in your terminal."

1.2 Vercel

Validate: vercel whoami
Fail: "Run `vercel login`"

1.3 Railway

Validate: railway whoami
Fail: "Run `railway login`"

1.4 Convex

Validate: ls ~/.convex/credentials 2>/dev/null || npx convex auth check
Fail: "Run `npx convex login`"

1.5 Cubic

Prompt: "Confirm Cubic GitHub App is installed on your account (cubic.dev)"

1.6 Slack

Prompt: "Where should investor updates go? (1) DM (2) #channel (3) group DM"
Validate: send test message via slack plugin
Store: slack_destination + slack_target -> .harness/config.yml

1.7 Figma

Validate: mcp__figma__whoami
Fail: "Reconnect via Figma plugin"

1.8 Stripe

Skip unless business_model requires payments. If needed:

Validate: stripe config --list 2>/dev/null
Fail: "Run `stripe login`"

1.9 Sentry

Prompt: "Create a Sentry project at sentry.io. We'll connect via MCP."
Store: SENTRY_DSN -> .harness/secrets.env

1.10 PostHog

Prompt: "Create a PostHog project at posthog.com (free tier is fine)."
Store: NEXT_PUBLIC_POSTHOG_KEY, NEXT_PUBLIC_POSTHOG_HOST -> .harness/secrets.env

1.11 Domain Registrar

Only if domain is confirmed from Q13.

Prompt: "Point your domain DNS to Vercel. Add these records: ..."

1.12 Persist

Write .harness/secrets.env (gitignored). Write .harness/config.yml with all connections. Print connection summary table.

Slack: "All services connected. {N}/{M} configured. Starting market research."

---

PHASE 2 — Research & Strategy

Agent: Commander + WebSearch | Supporting: Writing (reports)

2.1 Competitor Research

Use WebSearch. Search for:

• "{keywords}" startup / "{keywords}" app / "{keywords}" SaaS
• Competitors on Product Hunt, HN, Reddit, G2
• Design patterns competitors use

Per competitor: name, URL, features, pricing, strengths, weaknesses.

2.2 Market Sizing

• TAM/SAM/SOM estimates with sources
• Growth rate of the sector
• Adjacent markets

2.3 Target Audience Validation

• Validate the target users from Q3 against competitor user bases
• Identify underserved segments
• Pain points from Reddit, HN, Twitter discussions

2.4 Positioning Strategy

• Key differentiator (what competitors miss)
• Positioning statement (for {target}, who {need}, {product} is a {category} that {benefit})
• Pricing strategy benchmarked against competitors

2.5 Business Model Validation

• Unit economics projections (CAC, LTV) based on comparable companies
• Pricing tiers recommendation
• Revenue model viability assessment

2.6 Compliance Research

If compliance[] is non-empty:

• Specific requirements for each compliance framework
• Technical implications (encryption, audit logs, data residency)
• Timeline/cost estimates for certification

Output: Write research-report.md — executive summary, competitor table, market sizing, audience profile, positioning, compliance notes. Create GitHub Issue: "Research: Market Analysis" with executive summary.

Slack: "Research complete. {N} competitors analyzed. Key differentiator: {one-liner}. Market size: {TAM}. Target: {audience}."

---

PHASE 3 — Product Spec

Agent: Commander | Reads: research-report.md, founder-profile.yml, .harness/stacks.yml

3.1 Core Spec

Generate product-spec.md with:

• Product overview + value prop + persona
• Pages/routes (URL, purpose, components)
• Features grouped by page with acceptance criteria (machine-verifiable)
• Priority: P0 (MVP, must ship), P1 (important, next sprint), P2 (nice-to-have)

3.2 Data Model

• Convex schema (tables, fields, indexes, relationships)
• Data flow diagrams

3.3 API Design

• API routes (method, path, request/response, auth requirements)
• Third-party integration points from Q10
• Webhook endpoints (if receiving external events)

3.4 Component Inventory

• Component list (name, props, variants, usage)
• Shared vs page-specific components

3.5 Startup-Type Adaptations

DevTool additions:

• CLI command structure
• SDK interface design (TypeScript, Python, Go)
• OpenAPI spec outline
• Documentation site structure (getting started, API reference, examples)

Hardware additions:

• Firmware architecture
• Communication protocol (BLE, WiFi, LoRa, Zigbee)
• OTA update mechanism
• Device provisioning flow

Marketplace additions:

• Supply-side onboarding flow
• Demand-side onboarding flow
• Trust & safety requirements
• Payment escrow / split payment design
• Review/rating system

Fintech additions:

• Transaction flow with reconciliation
• Audit logging requirements
• KYC/AML integration points

Healthcare additions:

• PHI data flow mapping
• Consent management
• Audit trail design
• BAA requirements per vendor

3.6 Compliance Spec

If compliance requirements exist: map each requirement to specific technical implementation (encryption at rest, field-level encryption, audit tables, consent flows).

3.7 GitHub Issues

Create GitHub Issues for every P0/P1 feature with acceptance criteria and labels.

gh project create --title "{Startup Name}" --owner @me

Add all issues. Columns: Backlog, In Progress, In Review, Done.

Slack: "Product spec complete. {N} features ({P0} critical, {P1} important). Top priorities: {top 5 P0 features}."

---

PHASE 4 — Design

Agent: Website (Figma) | Supporting: Writing (brand voice)

4.1 Brand Identity

• Name validation (domain availability check if not done)
• Color palette (saved as Tailwind tokens)
• Typography selection (NOT Inter)
• Logo placeholder
• Voice/tone guidelines (saved in SOUL.md or .harness/brand.yml)

4.2 UI/UX Design

If founder provided Figma designs (Q12): use those as source of truth. If not: generate designs.

For each page in product spec:

1. Use mcp__figma__generate_figma_design — all components, desktop + mobile
2. Save screenshot to .harness/design-screenshots/{page-name}.png

Requirements:

• Mobile responsive breakpoints
• Accessibility (WCAG 2.1 AA)
• Dark mode (if applicable)
• Empty states, loading states, error states

4.3 Design System

Write .harness/design-system.md — colors, typography, spacing, component tokens. Configure Tailwind v4 with these tokens.

4.4 Domain & Email Setup

If domain confirmed:

• Configure DNS
• Set up email (Resend with custom domain)
• SSL (auto via Vercel/Cloudflare)

Slack: "Designs complete. {N} pages in Figma. Brand identity established. {Figma URL if available}."

---

PHASE 5 — Infrastructure Setup

Agent: Ops | Supporting: Backend

5.1 Repository Scaffold

npx create-next-app@latest {project-dir} --typescript --tailwind --eslint --app --turbopack --import-alias "@/*"
cd {project-dir}
npm install @tanstack/react-query zustand convex
npm install -D vitest @vitejs/plugin-react playwright @playwright/test
npx convex init
git init && git add . && git commit -m "Scaffold: {name} with Next.js + Convex + Tailwind v4"
gh repo create {repo-name} --private --source=. --push

5.2 CI/CD Pipeline

Write .github/workflows/ci.yml: vitest + playwright on push, Cubic scan on PR. Configure: staging (auto-deploy on PR merge), production (deploy on release tag).

5.3 Deploy Infrastructure

• vercel link (frontend)
• railway link (backend, if needed)
• Convex project setup
• Environment variables in all services

5.4 Monitoring & Observability

• Sentry integration (via MCP + @sentry/nextjs)
• PostHog analytics setup
• Uptime monitoring (BetterUptime or OpenStatus)
• Lighthouse CI (performance baselines)

5.5 Startup-Type Infrastructure

DevTool additions:

• npm/PyPI publishing pipeline in CI
• Documentation site scaffold (Nextra, Mintlify, or Docusaurus)
• API playground / sandbox
• CLI distribution (Homebrew formula, npm global, curl installer)

Hardware additions:

• Firmware CI/CD pipeline (PlatformIO, Arduino CLI)
• OTA update server endpoint
• Device provisioning / fleet management dashboard
• BOM tracking spreadsheet or database table

Marketplace additions:

• Separate onboarding flows for supply + demand sides
• Stripe Connect for split payments

Fintech additions:

• PCI-compliant infrastructure configuration
• Encryption at rest for sensitive fields
• Audit log table + write-only API

Healthcare additions:

• HIPAA-compliant hosting configuration
• BAA tracking per vendor
• PHI encryption + access logging

5.6 Harness Configuration

• Set up .harness/ configs
• Register hooks (GateGuard, config-protection)
• Set up .mcp.json (Cubic, Sentry)
• Configure Taskfile

Slack: "Infrastructure ready. Repo: {GitHub URL}. CI/CD configured. Monitoring active. Ready to build."

---

PHASE 6 — Build (TDD Loop)

Agents: Website + Backend | Supporting: Ops (CI), Growth (analytics hooks)

6.1 Test Suite (Red Phase)

Write tests BEFORE implementation for every P0 feature:

• Unit (Vitest): acceptance criteria, edge cases, error states
• E2E (Playwright): user flows, navigation, forms, visual comparison vs design screenshots

npx vitest run --reporter=verbose   # ALL red
npx playwright test --reporter=list  # ALL red

Commit tests separately: git commit -m "Tests: failing tests for {feature list}"

6.2 Per-Feature Build Cycle

For each P0 feature in dependency order:

1. Implement — Read acceptance criteria from Issue. Read Figma via mcp__figma__get_design_context. Build it. Follow .harness/design-system.md.
2. Gate 1: Tests — npx vitest run && npx playwright test. Fix until green. Max 10 iterations.
3. Gate 2: Cubic — Create PR, push, get Cubic review. Fix issues. Max 5 loops.
4. Gate 3: Visual QA — npx playwright test --grep @visual. Compare vs design screenshots. Diff > 2%? Fix. Max 5 loops.
5. Ship — All gates green: gh pr merge --squash. Move Issue to Done.

Slack per feature: "Feature shipped: {name}. Tests: {N} passing. Cubic: clean. Progress: {done}/{total}."

6.3 Backend Build (if applicable)

• Convex schema + functions
• API routes
• Authentication (Clerk/Better Auth when ready)
• Authorization / roles
• Rate limiting (Upstash)
• Webhook receivers

6.4 Integration Build

• Payment processing (Stripe)
• Email sending (Resend)
• File uploads (Uploadthing)
• Search, maps, messaging per Q10
• Domain-specific APIs

6.5 DevTool-Specific Build

• CLI implementation
• SDK client libraries
• API documentation (OpenAPI spec)
• Getting started guide + code examples

Context Reset Protocol

Context filling? Commit everything. Update all Issues with status. Next session:

gh issue list --state open --label "priority:P0" --json number,title,body,comments
gh project item-list --owner @me --format json

---

PHASE 7 — Launch Preparation

Agent: Commander (coordinates all agents to verify their domain)

7.1 Legal

• Terms of Service (generated from template, adapted to startup type)
• Privacy Policy (adapted to compliance requirements)
• Cookie consent (if EU users / GDPR)
• GDPR data export/deletion endpoints (if EU)
• Any domain-specific legal (HIPAA notices, financial disclaimers)

7.2 SEO

• Sitemap generation (sitemap.xml)
• robots.txt
• Meta tags on all pages (title, description, OG images)
• Structured data / JSON-LD where applicable
• Social media preview cards

7.3 Analytics Verification

• PostHog events firing correctly
• Key funnels configured (signup, activation, conversion)
• Feature flags ready for experiments

7.4 Error Tracking Verification

• Trigger test error, confirm Sentry captures it
• Source maps uploaded
• Alert rules configured

7.5 Pre-Launch Checklist

• [ ] All P0 features complete and tested
• [ ] Cubic codebase scan clean
• [ ] Performance: Lighthouse score > 90
• [ ] Accessibility: axe-core audit passed
• [ ] Security: no exposed secrets, rate limiting enabled
• [ ] Monitoring: Sentry + uptime + PostHog all verified
• [ ] Legal: ToS + Privacy Policy published
• [ ] SEO: sitemap + robots.txt + meta tags
• [ ] Backup strategy configured
• [ ] Support channel set up (email, Discord, or Intercom)

7.6 Staging Validation

• Full e2e test suite on staging
• Payment flow test (Stripe test mode, if applicable)
• Manual smoke test (Playwright walks core flows)

Slack: "Launch checklist complete. {N}/{M} items verified. Ready to deploy."

---

PHASE 8 — Deploy & Go Live

Agent: Ops | Supporting: All agents verify their domain

vercel --prod                          # Frontend
npx convex deploy                      # Database
railway up                             # Backend (if any)
curl -s -o /dev/null -w "%{http_code}" {prod-url}  # Health check
PLAYWRIGHT_BASE_URL={prod-url} npx playwright test   # E2E against prod

Fail? Read logs, fix, retry. Max 5 attempts. Still failing? Slack full error log.

Slack:

DEPLOYED — {Name} is LIVE URL: {url} Features: {N} shipped. Tests: all passing. Cubic: clean. Stack: Next.js + Tailwind v4 + Convex on Vercel + Railway Next: {P1_count} features in backlog.

---

PHASE 9 — Growth (ongoing)

Agent: Growth | Supporting: Writing, Website (A/B tests)

9.1 Analytics & Metrics

• Daily/weekly metrics digest (PostHog data → Slack)
• Funnel analysis (signup → activation → conversion)
• Retention tracking
• Feature usage tracking

9.2 Content Marketing

• Blog posts (Writing agent, weekly/biweekly)
• Social media posts (Writing agent)
• SEO optimization (Growth agent monitors rankings)
• Landing page A/B testing

9.3 User Acquisition Channels

• Product Hunt launch (Writing agent drafts, human reviews)
• Hacker News "Show HN"
• Reddit posts in relevant subreddits
• Twitter/X threads
• LinkedIn posts
• Dev.to / Hashnode articles (if DevTool)
• YouTube demos / tutorials

9.4 Payments & Revenue (if applicable)

• Stripe dashboard monitoring
• Churn analysis
• Upgrade/downgrade tracking
• Tax compliance (Stripe Tax)

Slack (weekly): "Week {N}: {X} users, {Y} revenue, {Z} features shipped. Top metric: {A}. Focus: {B}."

---

PHASE 10 — Maintenance (ongoing, reactive)

Agent: Ops | Supporting: Website/Backend (fixes), Commander (escalation)

10.1 Error Response

Sentry MCP detects error → Ops agent investigates → fix → deploy → verify. Escalate to Slack if can't fix within budget.

10.2 Uptime Monitoring

Monitor down → webhook → Ops agent → diagnose → fix → verify. Status page updated automatically.

10.3 Dependency Updates

• Weekly npm audit
• Auto-update patches
• Flag breaking changes for human review

10.4 Performance Monitoring

• Weekly Lighthouse CI run
• Core Web Vitals regression alerts
• Database query performance

10.5 Security Scans

• Monthly Cubic codebase scan
• Dependency vulnerability scanning
• Secret rotation reminders
• SSL certificate monitoring

---

PHASE 11 — Iteration (ongoing, proactive)

Agent: Commander (prioritization) | Supporting: All agents

11.1 User Feedback Loop

• In-app feedback widget
• NPS surveys
• Feature request tracking (→ GitHub Issues)
• Bug reports (→ GitHub Issues → Ops agent)
• User feedback → prioritize → spec → build (back to Phase 6)

11.2 A/B Testing

• Growth experiments via PostHog feature flags
• Measure → keep/discard
• Report results with actual numbers

11.3 Platform Expansion

• Mobile app (React Native, Expo)
• Desktop app (Electron, Tauri)
• Browser extension
• API public release (if DevTool)
• Marketplace integrations (Zapier, Make)

11.4 Fundraising Support (if applicable)

• Pitch deck draft (Writing agent)
• Metrics dashboard for investors
• Due diligence document preparation

---

Agent Assignments Summary

| Phase | Primary Agent | Supporting Agents | |-------|--------------|-------------------| | 0 (Interview) | Commander | — | | 1 (Connections) | Commander | — | | 2 (Research) | Commander + WebSearch | Writing (reports) | | 3 (Spec) | Commander | — | | 4 (Design) | Website (Figma) | Writing (brand voice) | | 5 (Infrastructure) | Ops | Backend | | 6 (Build) | Website + Backend | Ops (CI), Growth (analytics) | | 7 (Launch Prep) | Commander | All agents verify | | 8 (Deploy) | Ops | All agents verify | | 9 (Growth) | Growth | Writing, Website | | 10 (Maintenance) | Ops | Website/Backend, Commander | | 11 (Iteration) | Commander | All agents |

---

Slack Update Templates

Phase complete: {Name} — {Phase Title} + 2-3 sentences + key metrics + next phase. Feature shipped: {Name} — Feature: {feature} + test/cubic/visual status + progress fraction. Blocker: {Name} — Attention Needed + issue + impact + plan. "Reply if you want a different approach." Weekly: {Name} — Week {N} + shipped + in progress + blockers + metrics + ETA. Milestone: {Name} — LIVE or {Name} — {X} Users + celebration + metrics.

---

Error Recovery

| Scenario | Action | |----------|--------| | Auth expired | Slack user, pause | | Tests loop > 10 | Slack details, skip feature, move on | | Cubic > 20 issues | Break PR into smaller pieces | | Figma MCP down | Build from spec + screenshots | | Deploy fails x5 | Slack full error log | | Context filling | Commit, update Issues, reset | | Rate limited | Exponential backoff + retry | | Compliance gap found | Slack immediately, do not ship without resolution |

---

Course Correction

User replies in Slack? Pick up via /slack:find-discussions.

• Feature change → update spec + Issues
• Design change → update Figma + re-run visual QA
• Priority change → reorder queue
• Stop → halt, post status
• New feature → add to spec + Issue + queue
• Pivot → restart from Phase 2 with new direction

Acknowledge: "Got it. {change}. Adjusting."