Kimchi — Development skill for Claude Code

🌶️ kimchi

**Turn a raw idea into build-ready docs your AI can just execute.**


**Part 1 of 2 — kimchi decides *what* to build.** **[🎅 SantaClaude](https://github.com/chitransh-cj/SantaClaude) builds it.** It picks up the EPIC docs and `execute.md` that kimchi emits and runs them as small, independently verifiable micro-tasks — dispatching each to a fresh subagent and auditing the work before it counts as done. Plan it here, ship it there.

Having ideas was never the problem. Getting your AI to *build* one correctly is. You open Claude or Codex, type a vague prompt, and burn hours re-explaining yourself while it fills gaps you never specified and builds the wrong thing — because the real spec only ever lived in your head. A few frustrating hours later the idea joins the graveyard.

**kimchi gives you the spec.** You hand it a raw idea; it hands back a clean, easy-to-read set of **build-ready docs**:

• One doc per EPIC — user stories with business severity/complexity and a strict priority order.
• Locked decisions — tech stack, architecture, exact API contracts, security and deployment calls, all fixed so a build session never re-litigates them.
• execute.md — a self-driving handoff that lets Claude Code or Codex build the whole thing story-by-story, autonomously, across multiple sessions, resuming from trackers.
• A net tracker + pre-build audit — whole-product status at a glance, and a gate that catches flawed decisions while they're still a doc edit, not a rewrite.

No more prompt-wrestling. You give the docs to your AI and it executes.

How it gets there: instead of one agreeable assistant, kimchi puts your idea through 11 world-class expert personas (product, architecture, design, security, finance, a deliberately lazy staff engineer, and more) that **grill you, counter you, and refuse to flatter you** until every decision is nailed down. The grilling is the en