Claude Arsenal

25 Claude Code subagents, 14 lifecycle skills, and CLAUDE.md templates — everything you need to build, ship, and monetize a product with AI, in one portable toolkit.

Specialist agents (backend, frontend, security, growth, legal, finance…) plus slash-command skills for every stage of a product's life, from idea validation to deployment, launch, and growth. This repo is the source of truth — portable for life: clone it on any machine, run sync.ps1.

One-time setup (per machine)

1. Clone this repo.
2. Run .\sync.ps1 — validates every agent/skill frontmatter (fails loudly on broken files), then copies agents/ and skills/ into ~/.claude/ and bundles templates/ into the deployed new-project skill.
3. Install the companion plugins (in Claude Code: /plugin): superpowers, frontend-design, playwright, supabase, context7. They provide the process layer (brainstorming, planning, TDD, debugging, design, browser testing, live docs) — this toolkit deliberately doesn't duplicate them.

After editing any agent/skill here, re-run .\sync.ps1.

The lifecycle: idea → income → company

1. Validate → /validate-idea — evidence and a go/no-go verdict before any code exists.
2. Idea → /new-project — brainstorms the design (via superpowers), scaffolds the folder, applies the right CLAUDE.md, git init, verifies it runs.
3. Money first → /monetize — customer, pricing, MVP-to-charge feature list. Do this before building features on anything meant to earn; pricing shapes scope.
4. Spec → /product-spec — PRD, flows, and a feasibility pass before building anything non-trivial.
5. Build → work in the project with Claude Code. Plans via superpowers:writing-plans; delegate bounded tasks to the builder agents; TDD via superpowers. Common features have skills: /add-auth, /add-ai.
6. Check → /code-review on the diff; code-reviewer / security-auditor agents for deeper passes; /harden before real users or payments; /optimize when it's slow or costly.
7. Ship → /ship-it — pre-flight checklist, deploy (Vercel + Supabase), verify the live URL.
8. Launch → /launch — assets, channel plan, drafted posts, day-of runbook.
9. Grow → /grow — numbers-first experiment loop, monthly or when growth stalls.
10. Run the business → /maintain (monthly upkeep), /pitch (fundraising), /hire (first contractors/employees).

Agents (agents/ → ~/.claude/agents/)

Agent	Model	Use for
architect	sonnet	System/feature design before building; returns design docs
backend-builder	sonnet	API routes, business logic, integrations (Stripe, webhooks)
frontend-builder	sonnet	Pages, components, forms, styling, accessibility
db-designer	sonnet	Schemas, migrations, RLS policies, indexes
test-engineer	sonnet	Tests, bug reproductions, edge-case hunting
code-reviewer	sonnet	Read-only diff review: bugs, security, complexity
security-auditor	sonnet	Read-only defensive audit: auth, RLS, secrets, input
devops-engineer	sonnet	Deploys, CI/CD, Dockerfiles, env var management
debugger	sonnet	Root-cause investigation with evidence
researcher	haiku	Library comparisons, docs research, broad exploration
docs-writer	haiku	READMEs, guides, API docs (verified against code)
biz-strategist	sonnet	Monetization, pricing, MVP scoping, landing-copy draft, launch
product-manager	sonnet	PRDs, user stories, scope cutting, prioritization
ux-designer	sonnet	User flows, screen maps, IA, onboarding, usability review
mobile-builder	sonnet	React Native/Expo screens, native features, store release
data-analyst	sonnet	Metrics/KPIs, analytics SQL, tracking plans, funnels/cohorts
ai-engineer	sonnet	Claude API features, prompts, RAG, evals, cost tuning
performance-engineer	sonnet	Profiling, Web Vitals, slow queries, caching — numbers first
refactor-engineer	sonnet	Behavior-preserving refactors, upgrades, migrations
automation-engineer	sonnet	Scripts, scrapers, pipelines, scheduled jobs, API glue
growth-marketer	sonnet	SEO, channels, email lifecycle, funnel diagnosis
copywriter	sonnet	Landing/email/microcopy, blog, launch posts, store listings
finance-analyst	sonnet	Unit economics, SaaS metrics, projections, runway
legal-advisor	sonnet	ToS/privacy, GDPR basics, OSS licenses, contractor checklists
market-analyst	sonnet	TAM/SAM/SOM sizing, competitive landscape, segments, why-now

Delegation habits (from global CLAUDE.md): don't spawn an agent for what two Grep/Read calls answer; parallelize independent agent tasks; builders get bounded tasks with acceptance criteria.

Skills (skills/ → ~/.claude/skills/)

Build:

• /validate-idea — evidence-based go/no-go before any code.
• /new-project — idea to running, git-tracked, Claude-ready skeleton (webapp / api / other presets).
• /product-spec — PRD + flows + feasibility gate, saved into the project.
• /add-auth — Supabase auth done right: RLS, SSR helpers, verified flow, audit.
• /add-ai — LLM features with evals, cost estimates, and injection guardrails.

Ship & harden:

• /ship-it — pre-flight, deploy, live verification.
• /harden — full security pass: audit → fix → verify → accepted risks.
• /optimize — measure → fix top bottlenecks → re-measure → perf budget.
• /maintain — monthly upkeep: errors, dep updates, backups, costs.

Earn & grow:

• /monetize — model, pricing, Stripe plan, landing page, launch checklist.
• /launch — launch assets, channel plan, drafted posts, day-of runbook.
• /grow — numbers-first growth loop: funnel → bottleneck → one experiment.
• /pitch — investor narrative, sourced numbers, deck outline, objection prep.
• /hire — role scorecard, JD, sourcing, structured interview loop, contract checklist.

Templates (templates/ — used by /new-project)

sync.ps1 bundles this folder into ~/.claude/skills/new-project/templates/ so the deployed skill is self-contained. Edit templates here, never the synced copy.

• CLAUDE.webapp.md — Next.js + TypeScript + Tailwind + Supabase + Vercel (the default stack)
• CLAUDE.api.md — Python + FastAPI
• CLAUDE.generic.md — anything else
• project-structures.md — reference layouts: CLI, extension, desktop (Tauri), mobile (Expo), static site, monorepo

Extending the toolkit

• New agent: add agents/<name>.md (frontmatter: name, description starting "Use this agent...", model, optional tools) — keep haiku for research-shaped work, sonnet for building. Re-sync.
• New skill: add skills/<name>/SKILL.md (frontmatter: name, description starting "Use when..."). The superpowers:writing-skills skill guides good skill design. Re-sync.
• New stack preset: add a templates/CLAUDE.<type>.md and a scaffold section in skills/new-project/SKILL.md.
• Keep everything here versioned — commit after each change so no machine ever holds the only copy.

Accounts this toolkit assumes (create once)

GitHub (code + CI), Vercel (hosting), Supabase (DB/auth), Stripe (payments), a domain registrar. All have free tiers except the domain.

License

MIT — see LICENSE. Use it, fork it, build on it.

---

If this toolkit saves you time, a ⭐ helps other builders find it.