Agentic-first prompt workflows. Markdown templates that teach AI agents how to optimize prompts, create PRDs, and manage implementation. Works with Claude Code, Cursor, Windsurf, GitHub Copilot, and many AI coding tools.

Your command format depends on your AI tool:

Rule of thumb: CLI tools use colon, IDE extensions use hyphen.

npm install -g clavix
clavix init

During clavix init, pick your AI tools (Cursor, Claude Code, GitHub Copilot, etc.). Clavix installs slash command templates into those integrations.

# CLI-style tools (Claude Code, Gemini, Qwen)
/clavix:improve "Create a secure login page with JWT"

# IDE-style tools (Cursor, Windsurf, GitHub Copilot, etc.)
/clavix-improve "Create a secure login page with JWT"

The AI agent reads the Clavix template and optimizes your prompt.

graph TD
    A["/clavix:improve"] --> B["/clavix:implement"]
    C["/clavix:prd"] --> D["/clavix:plan"]
    D --> E["/clavix:implement"]
    E --> F["/clavix:verify"]
    F --> G["/clavix:archive"]
    H["/clavix:start"] --> I["/clavix:summarize"]
    I --> D
    J["/clavix:refine"] --> D
    J --> B

/clavix:improve → /clavix:implement

Optimize a prompt and implement it directly.

/clavix:prd → /clavix:plan → /clavix:implement → /clavix:verify → /clavix:archive

Structured planning with PRD, task breakdown, implementation, verification, and archival.

/clavix:start → [conversation] → /clavix:summarize → /clavix:plan

Have a conversation to explore requirements, then extract and plan.

/clavix:refine → (updated PRD or prompt) → continue workflow

Refine existing PRDs or prompts based on feedback.

/clavix:review is for reviewing teammates' PRs (criteria-driven, severity levels, saved reports), while /clavix:verify checks your implementation against your own PRD.

For a full walkthrough, see Getting Started.

Clavix is agentic-first:

1. You run clavix init – Select integrations (Cursor, Claude Code, GitHub Copilot, etc.). Clavix installs markdown templates into each tool (e.g. .cursor/commands/, .github/prompts/clavix-*.prompt.md).
2. You invoke a slash command – Like /clavix:improve, /clavix:plan, or /clavix:review in your AI tool.
3. The AI agent reads the template – Structured markdown instructions with frontmatter.
4. The agent follows the instructions – Using its native tools (edit files, run tests, summarize, review PRs, etc.).
5. Outputs are saved locally – Under .clavix/outputs/ (including .clavix/outputs/reviews/ for PR review reports).

No TypeScript executes during slash commands. The markdown templates ARE the product; Clavix only installs and updates them.

See Architecture for details.

In GitHub Copilot Chat, Clavix commands appear as /clavix-improve, /clavix-prd, /clavix-review, etc.

Full list: docs/integrations.md

All workflows (/clavix:improve, /clavix:plan, /clavix:review, etc.) are slash commands that your AI tools execute using Clavix templates.

• Getting Started - Installation and first workflow
• Commands Reference - All commands in one place
• Architecture - How Clavix works
• Integrations - Full tool matrix
• CONTRIBUTING.md - Contribute to Clavix

• Node.js >= 18.0.0
• npm or yarn
• An AI coding tool (Claude Code, Cursor, GitHub Copilot, etc.)

Apache-2.0