---
slug: "rust-agentic-skills"
source_type: "readme"
source_url: "https://cdn.jsdelivr.net/gh/udapy/rust-agentic-skills@main/README.md"
repo: "https://github.com/udapy/rust-agentic-skills"
source_file: "README.md"
branch: "main"
---
# Rust Agentic Skills

[![Rust Guild](https://img.shields.io/badge/Agent-Rust%20Guild-orange)](https://github.com/rust-lang/rust)
[![Research-Plan-Implement Methodology](https://img.shields.io/badge/Methodology-RPI-blue)](AGENTS.md)
[![Claude Ready](https://img.shields.io/badge/Claude-Plugin-purple)](.Claude-plugin/marketplace.json)
[![Gemini Extension](https://img.shields.io/badge/Gemini-Extension-blue)](gemini-extension.json)
[![Antigravity Compatible](https://img.shields.io/badge/Antigravity-Engine-green)](AGENTS.md)
[![Install with Skills.sh](https://img.shields.io/badge/Install_with-skills.sh-000000?style=for-the-badge)](https://skills.sh)

**A modular, constraint-based skill set for Autonomous AI coding agents.**

This repository transforms any general-purpose LLM (Claude, Gemini) into a **disciplined Rust engineering team**. It adheres to the **Agent Context Protocol (ACP)** to provide self-describing skills that explicitly define their triggers, capabilities, and execution phases.

---

## // Architecture Overview

![Architecture Diagram](https://github.com/udapy/rust-agentic-skills/raw/HEAD/assets/rust-agentic-skill-architecture.png)

We do not use monolithic instruction files. Instead, every skill in `skills/` follows the **Brain-Tool-Context** architecture to maximize token efficiency:

1.  **Driver (`SKILL.md`)**:
    - **The Brain**. Contains minimal YAML metadata (`trigger`, `rpi_phase`) and high-level role benchmarks.
    - _Usage_: The agent reads this first to decide if it is relevant.
2.  **Tools (`scripts/`)**:
    - **The Hands**. Executable code (Shell scripts, Rust binaries) for reliable, deterministic task execution.
    - _Usage_: The agent executes these to perform work (e.g., `init_project.sh`, `explain_error.sh`).
3.  **Context (`references/`)**:
    - **The Knowledge**. Deep-dive documentation and "Dictionaries of Pain".
    - _Usage_: "Lazy loaded" by the agent only when specifically needed to solve a complex problem.

---

## // Integration Guide

### 1. Claude Code (Plugin)

This repository is a compliant Claude Plugin.

1.  Clone this repository locally.
2.  Allow Claude to discover the `.Claude-plugin/marketplace.json` manifest.
3.  **Result**: Claude will automatically see "Rust Kernel" and "Lint Hunter" as available tools.

### 2. Gemini CLI (Extension)

Compatible with the Gemini CLI tool via the **Model Context Protocol (MCP)**. This repo provides a dynamic server that auto-discovers skills.

1.  **Link Extension**:
    ```bash
    gemini extensions link .
    ```
2.  **Activate**:
    ```bash
    gemini context set rust-agentic-skills
    ```
3.  **Result**: The `rust-agentic-skills` server starts up, reads your `skills/` directory, and dynamically routes your requests to the appropriate skill (e.g., "Lint Hunter").

### 3. Manual / Custom Agents

**Running the MCP Server Manually:**
You can run the server directly to verify standard JSON-RPC communication:

```bash
./bootstrap.sh
```

_(Or `cargo run --release` if you prefer manual builds)_

_(Expects JSON-RPC messages on stdin)_

For generic agents (ChatGPT, heavily customized setups):

1.  **System Prompt**: Load [AGENTS.md](https://github.com/udapy/rust-agentic-skills/blob/HEAD/AGENTS.md) as your system instruction. It defines the **RPI (Research → Plan → Implement)** loop.
2.  **Context Loading**: When the agent enters a specific phase (e.g., "Verification"), manually load the relevant `SKILL.md` (e.g., `skills/lint-hunter/SKILL.md`).

---

## // See it in Action

Want to see what a full session looks like?

- **[See Example Trace](https://github.com/udapy/rust-agentic-skills/blob/HEAD/examples/gemini_session_trace.md)**: A real-world interaction log showing the agent initializing a project, running commands, and validating outputs.

---

## // How to Contribute 🤝

We welcome new skills! Follow the **Triad Pattern**:

1.  **Create Directory**: `skills/<your-skill-name>/`.
2.  **Create Driver**: Add `SKILL.md` with YAML frontmatter:
    ```yaml
    ---
    name: My Skill
    description: What it does.
    rpi_phase: Implementation
    trigger:
      - "keyword1"
      - "keyword2"
    capabilities:
      - capability 1
    ---
    ```
3.  **Add Tools**: Put executable scripts in `skills/<your-skill-name>/scripts/`.
4.  **Add Context**: Put documentation in `skills/<your-skill-name>/references/`.
5.  **Generate Docs**: Run `make doc` to update `AGENTS.md`.

---

## // Roadmap (In Progress / TBD)

- [ ] **Multi-Agent Beast Mode**: Chaining multiple skills in a single "Beast Mode" loop without human intervention.
- [ ] **New Skill**: `Test Architect` (Refactoring & property-based testing).
- [ ] **New Skill**: `Crates.io Scout` (Dependency analysis).
- [ ] **Automated CI**: GitHub Action to run `make verify` on PRs.

---

## 📜 License

MIT