Kimotep Skills — Sistema de Prompts Estruturados para IA
Kimotep/skills
Prompts estruturados para design agentic, briefs e planejamento antes de codificar
Ver no GitHubBiblioteca de skills pré-codificadas que implementam fluxos agentic para Claude: design de sistemas, entrevistas socráticas e scaffolding de handover. Inclui 6 skills especializadas (context-compass, keen-eyeos, socratic-product, voice-compass, manifest, writer-voice) que guiam você com perguntas estruturadas antes da implementação, entregando briefs, especificações e planos prontos.
Instalação
npx convez-ai Kimotep/skills1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Scoping e planejamento de projetos de software antes de implementar
- Design de sistemas multi-agent com planos de falha e contracts de agentes
- Especificação de interfaces nativas iOS/macOS com componentes HIG-aligned
- Desenvolvimento de voz e tom para escritores e marcas
- Criação de handover docs e scaffolding de repositórios
Veredito

Skillset bem estruturado que otimiza a fase pré-desenvolvimento através de entrevistas guiadas e saída semiestruturada. Útil para scoping e planejamento agentic, mas requer disciplina do usuário para extrair valor máximo — não é automação, é scaffolding iterativo.
README
A tradução para o português deste README ainda está na fila.
Kimotep Skills
A curated library of AI prompt systems — structured prompts that help you think clearly before you design, build, plan, or write.
Each one is a self-contained system prompt (SKILL.md) — paste it into ChatGPT, Claude, Gemini, or any AI assistant that takes custom instructions. Instead of guessing what to ask, it asks you — in the right order — then hands back something concrete: a brief, a design spec, a project plan, or a new writer skill in your own voice.
If you're using Claude specifically, each folder can also be installed directly as a Claude Skill — see Quick start.
See it in action
You say: "I want to build a habit tracker app but I'm not sure what to focus on first." → context-compass asks a few quick questions (what's already decided, what you haven't thought about, where this is going) → You get: a ready-to-run brief — goal, context, one concrete first step — in under five minutes.
Who this is for
- Solo builders & indie hackers who want to scope an idea before opening an editor
- Apple platform developers & designers who want native-feeling SwiftUI/UIKit UI without guessing
- Founders & PMs planning multi-agent or AI workflow systems before writing any code
- Writers, marketers, and founders who want an AI that writes in their voice, not a generic one
- Anyone mid-session who wants to turn what they're working on into a polished, shareable PDF — without re-explaining the content
Quick start
- Pick a skill below and open its folder.
- Either:
- copy the contents of its
SKILL.mdand paste it as a system prompt or custom instructions in your AI assistant of choice (ChatGPT, Claude, Gemini, etc.), or - if you're using Claude: install the folder directly as a Skill in Claude Code, Cowork, or Claude Desktop.
- copy the contents of its
- Describe what you're working on — the skill takes it from there with guided questions.
No setup, no dependencies, no API keys.
Skills
| Skill | Use it when... | Produces |
|---|---|---|
| context-compass | You have a rough idea and want to sharpen it into a clear brief in a few minutes | A ready-to-run brief — or jumps straight into the work if you say "start now" |
| keen-eyeos | You're designing for iOS, macOS, iPadOS, watchOS, or visionOS and want native-feeling UI, component picks, or HIG-aligned screens | Design plan, wireframes, dev handover spec, device mockup |
| socratic-product (socratic-agentic-workflow) | You have an idea for an AI agent or multi-agent system and want to plan it before writing code | An indexed plan: mission, agent specs with contracts and failure paths, a starter repo scaffold, system logic, a handover doc — plus a harness root config file (CLAUDE.md, .cursorrules, etc.) ready to drop in the repo |
| voice-compass | You want an AI writer that sounds like you — for a blog, brand, or your own voice | An installable writer skill: voice profile + per-format reference docs |
| manifest | You want to give the current work a shareable form — say "manifest this" and the skill reads the room and decides the rest, or add intent ("as slides", "one-pager for the board") and it executes against that | A designed PDF: report, slides, data, or mixed — light, never dense |
| feature-compass | You're adding a new feature to an existing project and want to align on exactly what it is before building starts | A [feature]-BRIEF.md — agreed feature summary, task list, and a ready-to-paste coding session prompt |
Real-world examples
- "I have a vague idea for a side project and don't know where to start." →
context-compassturns it into a brief in under five minutes, or starts building right away. - "My SwiftUI settings screen feels off, but I can't say why." →
keen-eyeosdiagnoses the issue and recommends the native fix, with platform-aware reasoning. - "I want to build a multi-agent research tool but don't want to start coding blind." →
socratic-productruns a 30-minute structured interview (all multiple-choice) and hands you an indexed plan plus a ready-to-drop harness config file (CLAUDE.md,.cursorrules, etc.) for your coding tool. - "I want my AI to write newsletter drafts that actually sound like me." →
voice-compassinterviews you (and reads your writing samples) and outputs a new, installable "writes like you" skill. - "Manifest this." →
manifestreads the current conversation, decides what form the work should take, and produces a designed document — report, slides, or data — in one call. Or add intent: "as slides for the board" and it executes against that. - "I want to add a search feature to my app." →
feature-compassinterviews you (always multiple-choice), reads your project config, and hands back a brief with task list and a ready-to-paste coding session prompt.
Repo structure
.
├── context-compass/ # Quick idea-sharpening warm-up
├── keen-eyeos/ # Apple platform design companion
├── socratic-product/ # Agentic system planning interview
├── voice-compass/ # Personalized writer skill builder
├── manifest/ # Turn current work into a polished PDF
├── feature-compass/ # Scope and align on a new feature before building
├── LICENSE
└── README.md
Each skill folder follows the same pattern: SKILL.md (entry point — paste as a system prompt anywhere, or install directly if you're using Claude), README.md (full docs), CHANGELOG.md (version history), and a references/ or scripts/ folder where needed.
Shared design pattern
socratic-product and voice-compass share an operating pattern — a time-boxed Socratic interview, always-ask-never-assume, assumption classification, and pause/resume via a STATE.md checkpoint. One plans agentic systems; the other builds personalized writers.
Status
All skills are MIT-licensed and actively maintained. See each skill's CHANGELOG.md for full version history.
| Skill | Version |
|---|---|
| context-compass | 0.2.0 |
| keen-eyeos | 1.0 |
| socratic-product | 0.6.0 |
| voice-compass | 0.3.0 |
| manifest | 0.2.0 |
| feature-compass | 0.2.0 |
Author
Kim Tumaini Jørgensen — tumaini.dk
Feedback and issues welcome via tumaini.dk or GitHub Issues.
MIT License — free to use, adapt, and share.