dotnetclaw

<p align="center">
    <img width="748" height="163" alt="image" src="https://github.com/user-attachments/assets/d20230dc-ff45-4ad1-81b8-17153eed0922" />
</p>
<p align="center">
    <img width="300" height="300" alt="logo" src="https://github.com/user-attachments/assets/b4ea5571-3ff9-409a-8781-7335777c9acb" />
</p>

# 🦀 DotnetClaw

![.NET 10](https://img.shields.io/badge/.NET-10-blueviolet)
![License: MIT](https://img.shields.io/badge/License-MIT-green)
![Semantic Kernel](https://img.shields.io/badge/Semantic%20Kernel-✓-blue)

🦀 DotnetClaw is a personal AI assistant developed in .NET framework — Semantic Kernel agents, Telegram Bot channel, Agent Skills, MCP client, Playwright browser, Cursor agent CLI integration

## What is DotnetClaw?

DotnetClaw is a self-hosted AI coding assistant that runs on your machine.
It combines **Microsoft Semantic Kernel**, **MCP (Model Context Protocol)**,
a **Telegram bot**, and a **Playwright browser** into a single .NET 10 CLI agent
that can read/write files, run shell commands, browse the web, and control your coding 
agent like Cursor or Claude Code — all from natural language.

**Supports:** OpenAI · Azure OpenAI · Anthropic Claude

## Inspiration

DotnetClaw is inspired by 🦞 [OpenClaw](https://github.com/openclaw/openclaw) (formerly Clawdbot / Moltbot)
by [Peter Steinberger](https://twitter.com/steipete) — a personal AI assistant concept
that proved how powerful a well-wired agent loop can be. This is a .NET reimagining of that idea.

## Projects

### DotnetClaw CLI
- Slash commands (/status /ask /plan /agent /cursor /claude)
- Chat and control your DotnetClaw agent from terminal
- Show chat history
- Install agent skills from DotnetClawHub (hub)

> DotnetClaw CLI running in Windows Powershell Terminal:
<img width="1109" height="332" alt="image" src="https://github.com/user-attachments/assets/8b5f45f2-4111-45d7-bd99-45b9549e778d" />

> DotnetClaw CLI running in Ubuntu bash terminal (via WSL): 
<img width="1110" height="334" alt="image" src="https://github.com/user-attachments/assets/fc9b4205-2761-4735-b80b-5ea62def1827" />

### DotnetClaw Web UI 
- Onboarding
- Dashboard
- Chat
- Skills
- Terminal UI
- Telegram channel configuration

> Web UI for DotnetClaw developed in Blazor:
<img width="1852" height="778" alt="Screenshot 2026-03-26 143759" src="https://github.com/user-attachments/assets/48686145-1936-466e-bb2c-cbffe6764d15" />

### DotnetClawHub - Hub for Agent Skills
- Create & Publish own custom (agent) skill
- Browse agent skills
- Install agent skill via API
- SKILL.md support

> Agent Skills Hub for DotnetClaw inspired by ClawHub developed in Blazor:
<img width="1288" height="521" alt="image" src="https://github.com/user-attachments/assets/2a76e17e-4efe-4d66-8b03-9de723588546" />

> Browse & search for skills:
<img width="1281" height="596" alt="image" src="https://github.com/user-attachments/assets/f804389f-7a83-4009-90ae-0b6490897069" />

> Skill details & installation:
<img width="1010" height="577" alt="image" src="https://github.com/user-attachments/assets/4f389f7c-3620-4912-8ce9-31f33fedc246" />

### Prerequisites
- [.NET 10 SDK](https://dotnet.microsoft.com/download/dotnet/10.0)
- An OpenAI API key (or Azure OpenAI / Anthropic — see providers)

### Run

```bash
# Set your API key
set OPENAI_API_KEY=sk-...

# Restore & run
cd src/DotnetClaw
dotnet run
```

### Configuration

| Environment Variable        | Description                          | Default  |
|-----------------------------|--------------------------------------|----------|
| `DOTNETCLAW_PROVIDER`       | `openai` \| `azure` \| `anthropic`   | `openai` |
| `OPENAI_API_KEY`            | OpenAI API key                       | —        |
| `AZURE_OPENAI_ENDPOINT`     | Azure OpenAI endpoint URL            | —        |
| `AZURE_OPENAI_API_KEY`      | Azure OpenAI key                     | —        |
| `AZURE_OPENAI_DEPLOYMENT`   | Azure deployment name                | Model ID |

Or edit `appsettings.json` directly.

### REPL Commands

| Command   | Action                          |
|-----------|---------------------------------|
| `help`    | Show available commands         |
| `reset`   | Clear conversation history      |
| `history` | Print full conversation history |
| `exit`    | Quit                            |

### Run Tests

```bash
cd tests/DotnetClaw.Tests
dotnet test
```

## Agentic Loop

```
User Input
    │
    ▼
ClawAgentLoop.RunTurnAsync()
    │
    ├──► ChatCompletionAgent (SK)
    │        │
    │        ├── FunctionChoiceBehavior.Auto()  ← auto tool-call selection
    │        │
    │        ├── [Tool Call] Shell.run_command
    │        ├── [Tool Call] FileSystem.read_file
    │        ├── [Tool Call] Dotnet.find_csharp_projects
    │        │        ▲
    │        │        └── results fed back into context
    │        │
    │        └── Final text response  ◄── streamed to terminal
    │
    └──► Max iterations guard (default: 20)
```

## Workspace Identity Documents

On every session start (and on `reset`), DotnetClaw loads `*.md` files from `./workspace/`
and injects them into the system prompt before any user message is sent.

```
./workspace/
├── SOUL.md      ← Who the agent is (personality, values, style)
├── AGENTS.md    ← How it uses tools and handles multi-agent flows
├── USER.md      ← Who you are (role, prefs, tech stack)
├── CONTEXT.md   ← What you're working on right now
├── MEMORY.md    ← Persistent facts you want remembered
└── *.md         ← Any additional documents, loaded alphabetically
```

**Loading order** is controlled by `WorkspaceDocumentPriority` in `appsettings.json`
(default: `SOUL → AGENTS → USER → CONTEXT → MEMORY → TOOLS → RULES`).
Remaining `*.md` files follow alphabetically.

The workspace folder is **optional** — if it doesn't exist, DotnetClaw starts with the
base system prompt only.

### REPL commands

| Command          | Action                                          |
|------------------|-------------------------------------------------|
| `workspace`      | Show loaded documents table                     |
| `ws reload`      | Force reload from disk without resetting chat   |
| `prompt`         | Print the full effective system prompt          |

### Runtime skill

The agent can query workspace docs mid-conversation via the `Workspace` plugin:
- `list_workspace_docs` — table of loaded docs
- `get_workspace_doc SOUL` — fetch a specific doc's content
- `reload_workspace` — hot-reload from disk
- `get_workspace_context` — full injected context block

---

## Skills (Plugins)

### Shell
| Function              | Description                            |
|-----------------------|----------------------------------------|
| `run_command`         | Execute any shell command              |
| `list_directory`      | Tree-style directory listing           |
| `get_working_directory` | Return current working directory     |

### FileSystem
| Function       | Description                     |
|----------------|---------------------------------|
| `read_file`    | Read a text file                |
| `write_file`   | Create or overwrite a file      |
| `append_file`  | Append to an existing file      |
| `delete_file`  | Delete a file                   |
| `file_exists`  | Check if a path exists          |
| `find_files`   | Glob search for files           |

### Dotnet
| Function                  | Description                          |
|---------------------------|--------------------------------------|
| `find_csharp_projects`    | Locate .csproj files                 |
| `summarise_csharp_file`   | Structural summary of a .cs file     |
| `get_nuget_packages`      | List NuGet packages in a .csproj     |

### Workspace
| Function                  | Description                                        |
|---------------------------|----------------------------------------------------|
| `list_workspace_docs`     | Table of all loaded identity documents             |
| `get_workspace_doc`  

[truncated…]