Hey HN, I built this because I kept seeing the same 3 problems in every OpenClaw setup thread:

  1. JSON config errors (especially Windows backslash paths breaking everything)
  2. ECONNREFUSED — usually just an IPv6/localhost binding issue but nobody knows that
  3. No way to discover which of the 7,400+ community skills are actually worth installing

  ClawKit is a browser-based toolkit with four tools:

  - Config Wizard — generates validated JSON configs with one-click presets for OpenAI, DeepSeek, Anthropic, Ollama. Auto-escapes Windows paths. Uses Zod for real-time validation.
  - Local Doctor — npx clawkit-doctor@latest checks Node.js version, port conflicts, config file syntax, permissions. Zero install, read-only.
  - Cost Estimator — models exponential context growth across GPT-4.1, Claude, DeepSeek. Most people don't realize running an agent 24/7 can cost $300+/mo until they see the curve.
  - Skill Registry — searchable directory of all ClawdHub skills with install commands, config snippets, and source links.

  Everything runs client-side. API keys never leave the browser. No auth, no paywall.

  Stack: Next.js 15 (App Router + RSC), Tailwind, shadcn/ui, Prisma + Postgres for skill data, Fuse.js for search, Umami for privacy-first analytics. Deployed on Vercel.

  Source: https://github.com/branzoom/getclawkit-web

  I'm a solo developer. Happy to answer any questions about the architecture or the skill crawling pipeline.