Prompt Cause Marketplace
PhAlves23/promptcause-marketplace
Marketplace curado de skills de IA para Claude Code e agentes compatíveis
Ver no GitHubUm marketplace curado de AI skills para Claude Code e outros agentes de IA. Permite instalar capabilities focadas (skills) a partir de um catálogo centralizado, com revisão e versionamento seguro. Uma forma organizada de compartilhar e reutilizar prompts e workflows.
Instalação
npx convez-ai PhAlves23/promptcause-marketplace1 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
- Instalar skills de engenharia de prompts para melhorar workflows
- Gerenciar múltiplas capabilities de IA a partir de um único catálogo
- Compartilhar e descobrir skills verificadas para Claude Code
Veredito

Projeto bem estruturado que resolve um problema real: centralizar e versionär skills de IA de forma segura. A proposta é sólida (marketplace vs repo individual), mas ainda está em estágio inicial com apenas 1 skill no catálogo. Requer ecossistema robusto de contribuidores para virar realmente útil.
README
A tradução para o português deste README ainda está na fila.
Prompt Cause — Skill Marketplace
A curated catalog of AI skills for Claude Code and other agents. Add one marketplace, install any skill.
Works with Claude Code · GitHub Copilot · OpenAI Codex · Gemini CLI · OpenCode · Cursor · Windsurf · Cline
Table of contents
- What is this
- Quick start
- Catalog
- How it works
- Security model
- Using skills in other tools
- Contributing a skill
- FAQ
- Repository structure
- Roadmap
- License
What is this
An AI skill is a focused capability you can drop into an AI coding agent — a SKILL.md plus reference files that teach the agent to do one thing well, and to know when to do it.
Prompt Cause is a curated marketplace of those skills. Instead of hunting down individual repos and copying files around, you add one marketplace and install any skill from a single catalog — with a review-and-pin process that keeps third-party skills safe to run.
Why a marketplace instead of one repo per skill?
- One source for users. Add it once; the whole catalog is available.
- Granular. Install only the skills you want — nothing is bundled or forced.
- Open but curated. Anyone can submit a skill; every listing is reviewed and version-pinned before it ships.
Quick start
Claude Code
/plugin marketplace add PhAlves23/promptcause-marketplace
/plugin install <skill-name>
Restart Claude Code, then use the skill (e.g. say "improve this prompt" for prompt-engineering).
GitHub Copilot CLI
copilot plugin marketplace add PhAlves23/promptcause-marketplace
copilot plugin install <skill-name>@promptcause
For Cursor, Windsurf, Codex, Gemini, and OpenCode, each skill's own repo ships native install instructions — see the catalog and Using skills in other tools.
Catalog
| Skill | What it does | Category | Repo |
|---|---|---|---|
| prompt-engineering | Rewrites a raw draft into a production-grade prompt, with a changelog of what changed and why. Grounded in primary sources (Anthropic, OpenAI, Google, The Prompt Report). | productivity | prompt-engineering-skill |
Want your skill here? See Contributing.
How it works
Claude Code (and compatible agents) use a three-level hierarchy:
Marketplace → contains many Plugins → each Plugin contains one or more Skills
(this repo) (a catalog entry) (the actual SKILL.md)
This repository is the marketplace: a catalog defined in .claude-plugin/marketplace.json that points at skills. A skill can live in one of two places:
| Location | source in the catalog | Best for |
|---|---|---|
| Its own repo (recommended) | git-subdir / github / url, pinned to a sha | Skills maintained independently, incl. community ones |
| Inline in this repo | "./plugins/<name>" | Small skills donated to the catalog |
Either way, users add a single marketplace and install skills individually.
Security model
Listed skills run inside other people's agents, so safety is built into the process — not an afterthought. Three layers:
- Review before listing. Every submission is audited for hidden instructions (prompt injection, data exfiltration, covert tool/URL calls), unsafe scripts, and scope that doesn't match its description.
- Pinned commits. Externally-hosted skills are pinned to a reviewed commit
sha. A contributor can change their own repo freely, but those changes do not reach users here until a newshais reviewed and merged. - Re-review on update + CI enforcement. Updating a skill is a new PR changing the
sha(re-audited). Thevalidateworkflow rejects any external source that isn't pinned to a full commitsha.
The repo itself is hardened too: least-privilege CI permissions, GitHub Actions pinned by sha, CODEOWNERS review, and Dependabot. Full policy in SECURITY.md and CONTRIBUTING — Versioning & security.
Using skills in other tools
Skills here target Claude Code first, but the format is portable. Each skill's repo documents its own native install for:
| Tool | Mechanism |
|---|---|
| Claude Code / Copilot CLI | This marketplace (/plugin or copilot plugin) |
| OpenAI Codex | Native skill discovery (~/.agents/skills) |
| Gemini CLI | Extension |
| Cursor / Windsurf / Cline | Rules files (.cursor/rules, .windsurf/rules, .clinerules) |
| OpenCode | AGENTS.md / skills path |
Contributing a skill
Community skills are welcome. Two paths:
- Keep your skill in your own repo (recommended) — you own and maintain it; we list it (pinned to a reviewed
sha). - Contribute it inline — add it under
plugins/in this repo.
Start from skill-template/, then follow CONTRIBUTING.md. Every submission is reviewed for quality and security before listing.
FAQ
Every listing is reviewed and pinned to a specific reviewed commit, so post-review changes in a contributor's repo can't silently reach you. That said, a skill is third-party content — install skills you trust, and review the source repo if in doubt. No marketplace can guarantee third-party code is safe; this is defense-in-depth. See SECURITY.md.
/plugin update <skill-name> (Claude Code) or copilot plugin update. You get the latest reviewed version — i.e. whatever sha is currently merged in this catalog.
/plugin uninstall <skill-name> removes a skill. /plugin marketplace remove promptcause removes the catalog.
Yes — see Using skills in other tools. Each skill's repo has native instructions for Cursor, Windsurf, Codex, Gemini, OpenCode, and Cline.
No — and that's by design. Your listing is pinned to the sha that was reviewed. To ship an update, open a PR here changing the sha; it's re-reviewed before merge. This protects users from unreviewed changes. See CONTRIBUTING — Versioning & security.
Skills are designed to be lean: a small always-on footprint (the description + triggers), with heavier reference material loaded only when the skill actually fires. Check a skill's details (claude plugin details <name>) for its token cost.
Repository structure
promptcause-marketplace/
├── .claude-plugin/marketplace.json # the catalog (skills + pinned sources)
├── plugins/ # inline skills (Path B), if any
├── skill-template/ # copy this to start a new skill
├── scripts/bump_skill.py # advance owned skills to their latest sha
├── assets/ # banner (png + editable html source)
├── docs/ARCHITECTURE.md # how the catalog, pinning, CI and bump work
├── .github/
│ ├── workflows/validate.yml # JSON + catalog + sha-pin enforcement
│ ├── workflows/bump-own-skills.yml # opens a PR bumping owned skills
│ ├── ISSUE_TEMPLATE/submit_skill.yml
│ └── CODEOWNERS · dependabot.yml · PULL_REQUEST_TEMPLATE.md
├── CONTRIBUTING.md · SECURITY.md · CODE_OF_CONDUCT.md · CHANGELOG.md · LICENSE
Roadmap
- Grow the catalog with more first-party and community skills
- Per-skill category browsing as the catalog grows
- Automated source-resolves check in CI (verify each pinned
shaexists upstream) - A simple website listing the catalog
Have an idea? Open a discussion.
License
MIT for the catalog and tooling in this repo. Each externally-hosted skill keeps its own license.