kimchi
worthwhileStructured ideation-to-spec pipeline for vibe coders; real guardrails but trivially replicable by an engineer with a prompt chain.
What it is
A single SKILL.md file that, when installed as a Claude Code or Codex CLI plugin, chains 13 adversarial expert personas to interrogate a user's idea and produce structured product docs (EPICs, user stories, tech decisions, execute.md). Includes a pre-build audit gate.
How it differs from vanilla Claude
A vanilla Claude can simulate multiple personas with a single prompt and produce structured output. kimchi's delta is the curated 13-persona choreography, a resume-from-tracker mechanic, and the audit gate — all pre-assembled into a one-command skill instead of requiring the user to craft the chain of prompts themselves.
Skill, plugin, or workflow shift?
Skill: it's a single SKILL.md file placed in the agent's skills directory, invoked via `/kimchi`. No standalone binary, no API, no runtime — entirely inside the agent's context.
Devil's advocate — is this just complexity?
A capable Claude can already simulate multiple personas if asked: 'Act as product head, architect, security expert, and lazy engineer — interview me on this idea, challenge every assumption, then produce EPIC docs with user stories and a build order.' That's one paragraph, not 13 scripts. The audit gate is a pre-built checklist — useful but not technically beyond what a one-shot prompt produces. kimchi is a well-packaged prompt chain, not a new capability. For an engineer who already knows how to prompt, this is a time-saver at best; for a vibe coder, it's training wheels that may become a crutch.
What would make it better
Expose as a standalone CLI tool that works with any agent (Cursor, Copilot) and stores artifacts locally. Add a schema versioning system so docs survive agent model updates. Provide a minimal 'quick mode' that skips the full interview when the idea is already well-defined. Integrate with version control automatically (commit EPICs on generation).
The honest case for it
If you are a solo founder or non-technical creator who wants to stop fighting with AI prompts, kimchi works. It forces you to think decisions through before building, surfaces blind spots you'd miss, and hands your AI a spec it can execute without you hovering. The audit gate alone can save you from building on a flawed premise. It's a cheap, zero-config way to add discipline to your workflow.
Who it's for
Audience fit
Depth and leverage for a technical engineer who wants to understand it and level up their workflow — not just offload work.
Value for someone who wants a more capable tool without the technical depth — accessible, does-it-for-you.
AI engineers can write their own structured prompting pipeline; the value is for those who skip the planning step and want a structured spec on autopilot.