I read Nate Herk's "I Tried 100+ Claude Code Skills. These 6 Are The Best" today. He converged on 6 meta-tools (Skill Creator, Superpowers, GSD, Context Mode, Claude Mem, plus Frontend Design as a bonus) for the agency dev who builds automations for clients.

I converged on a different 6 — for the indie hacker who's shipping their own product. Same number, different layer. Here's what they are, and what the actual output of one looks like.

The 6

1. shipping-checklist — pre-deploy gate. Reads manifests, env vars, error handling, monitoring. Outputs a markdown checklist tied to actual file paths in your codebase.
2. launch-thread-writer — reads your repo + product page, drafts a 7-tweet launch thread that doesn't sound AI-generated.
3. support-reply-drafter — reads your docs + the inbound email + your last 3 replies, drafts a response in your tone.
4. pricing-page-generator — reads your product, your competitors, your audience. Outputs a pricing page draft + 3 alternative tiers.
5. architecture-decision-recorder — codifies a decision before you forget why you made it. Reads recent commits, asks 5 questions, outputs the ADR.
6. competitor-deep-dive — give it a competitor URL. Reads their landing, pricing, docs, recent posts. Outputs a 1-page intel brief + 3 wedges where you can win.

Replay over diagram

Diagrams of what skills do tell you the structure. They don't tell you whether the output is useful. Here's what shipping-checklist actually emits when you run it on a Fastify API repo:

# Shipping checklist — api-gateway

## Build
- [ ] tsconfig.json:14 — noEmit is true; production build needs tsc -p tsconfig.build.json (configured but not in package.json scripts)
- [x] package.json:23 — build script outputs to dist/

## Env
- [ ] src/config.ts:8 — DATABASE_URL accessed but not validated. Add zod check or fail-fast on boot.
- [ ] .env.example:11 — STRIPE_WEBHOOK_SECRET present in code (src/routes/webhook.ts:34) but missing from .env.example
- [x] src/lib/env.ts:1 — t3-env schema covers 8 of 9 required vars

## Errors
- [ ] src/routes/users.ts:42 — unhandled await on db.user.create() — wrap in try/catch and return 500 with request id
- [ ] src/middleware/error.ts:9 — error handler logs but doesn't include request id; trace correlation will fail in prod

## Monitoring
- [ ] no Sentry/observability config detected — add @sentry/node to package.json and init in src/index.ts before app.listen
- [x] src/index.ts:18 — health check endpoint at /healthz returns 200

## Deploy
- [ ] Dockerfile:11 — copies node_modules from local; should npm ci in builder stage to avoid host drift
- [x] vercel.json:3 — buildCommand and outputDirectory set

7 of 12 items pass · 5 blocking items remain · est. 35 min to clear

Every item points at a specific file:line. No "remember to set env vars" filler. That's the design rule across all six skills: code-grounded over template-driven.

Why this matters

If you're an agency dev building automations for clients, Nate's six (Skill Creator builds skills, GSD keeps context clean, etc.) are the right meta-tools. They're about process.

If you're the indie hacker building your own product, you need a different layer — skills that help you ship, launch, support, price, decide, compete. That's where the six above sit.

Free download

I made shipping-checklist (the smallest of the six) free, no email gate, no signup: agentstack-ecru.vercel.app/free. Drop the SKILL.md into ~/.claude/skills/shipping-checklist/ and it fires on phrases like "ready to ship" or "launch checklist".

The other five ship as the AgentStack Power Pack ($39, 14-day refund). They were built racing OpenAI's Codex to $10k in net profit on a $0 budget — public live race at agentstack-ecru.vercel.app/race.

Thanks @nateherk for the article — it was the prompt that articulated this layer.