Metadata-Version: 2.4
Name: antislope-ai
Version: 0.1.1
Summary: Local AI coding supervision layer — watches your code, runs on-device review, surfaces findings via MCP
License-Expression: Apache-2.0
Project-URL: Homepage, https://github.com/zcj220/antislope-ai
Project-URL: Repository, https://github.com/zcj220/antislope-ai
Project-URL: Issues, https://github.com/zcj220/antislope-ai/issues
Keywords: ai,code-review,mcp,ollama,local-ai,developer-tools
Classifier: Development Status :: 3 - Alpha
Classifier: Intended Audience :: Developers
Classifier: Programming Language :: Python :: 3
Classifier: Programming Language :: Python :: 3.11
Classifier: Programming Language :: Python :: 3.12
Classifier: Topic :: Software Development :: Quality Assurance
Requires-Python: >=3.11
Description-Content-Type: text/markdown
License-File: LICENSE
Dynamic: license-file

# antislope-ai

**Stop your vibe-coded project from turning into a mess.**

You're using Cursor, Copilot, or Claude Code to write code fast — but AI generates a lot of code quickly, and without a feedback loop, issues pile up silently until the codebase becomes too tangled to fix.

antislope-ai is a local background layer that:

- **Watches every save** → runs an on-device AI review automatically (no API cost, no code leaves your machine)
- **Catches issues early** → flags naming problems, missing docs, risky boundaries before they stack up into debt
- **Feeds your AI tool live context via MCP** → Cursor / Copilot / Claude Code automatically know your active rules, project structure, and recent findings without you pasting anything manually
- **Saves tokens** → instead of re-explaining your project in every conversation (2,000–20,000 tokens), MCP injects a compact, always-fresh summary (~1,000 tokens total)

> Designed for vibe coders and beginners: runs silently in the background, no manual code review needed, works with whatever AI coding tool you already use.

## How it works

```
  Your editor  ──saves──▶  watcher  ──triggers──▶  local model (Ollama, free)
                                                           │
                                                    review results
                                                           │
  AI coding tool  ◀──MCP tools──  dashboard (http://127.0.0.1:8771)
  (Copilot / Cursor / Claude Code)
```

- **Local model** (qwen2.5-coder:7b or any Ollama model) reviews every save against your project rules — runs on your machine, zero API cost
- **Dashboard** shows current issues, risk chains, and handling status
- **MCP endpoint** lets Cursor, VS Code Copilot, Claude Code, and others automatically read your active rules and recent findings — no manual copy-paste, fewer tokens per session

## Requirements

| Item | Version |
|------|---------|
| Python | ≥ 3.11 |
| [Ollama](https://ollama.com) | latest |
| macOS / Linux | — |

> Windows is not tested. Ollama runs on Windows but the shell commands differ.

## Quick start

```bash
# 1. Install Ollama and pull a model (one-time)
brew install ollama          # macOS
ollama pull qwen2.5-coder:7b

# 2. Clone and set up
git clone https://github.com/zcj220/antislope-ai.git
cd antislope-ai
python3 -m venv venv
source venv/bin/activate
pip install -e .

# 3. Initialize project
antislope init

# 4. Start the dashboard (also starts the MCP server)
antislope dashboard
# Open http://127.0.0.1:8771 in your browser
```

## Watch a file for live review

```bash
# In a second terminal
antislope watch --file path/to/your/file.py
```

## MCP integration (AI coding tools read your rules automatically)

The dashboard exposes an MCP endpoint at `http://127.0.0.1:8771/mcp` with four tools:

| Tool | Returns |
|------|---------|
| `get_active_rules` | Current active review rules |
| `get_project_structure` | Entry points and core file roles |
| `get_recent_issues` | Latest detected issues and risk level |
| `get_project_context` | Project goal, direction, and high-risk areas |

### Cursor

Config already included at `.cursor/mcp.json`. Restart Cursor — tools appear automatically.

### VS Code Copilot (Agent mode, v1.99+)

Config already included at `.vscode/mcp.json`. Restart VS Code → open Copilot Chat → switch to **Agent** mode → enable antislope tools.

### Claude Code

```bash
claude mcp add antislope http://127.0.0.1:8771/mcp
```

### Windsurf

Edit `~/.codeium/windsurf/mcp_config.json`:

```json
{
  "mcpServers": {
    "antislope": { "serverUrl": "http://127.0.0.1:8771/mcp" }
  }
}
```

## Other commands

```bash
antislope review-real --file path/to/file.py   # One-shot manual review
antislope stats                                 # Review event statistics
antislope validate-rules                        # Test rules against sample files
antislope clean-review-data                     # Normalize historical review data
antislope index-structure                       # Rebuild structure index
```

## Default model

The default model is `qwen2.5-coder:7b`. To change it, edit `data/model-settings.json`:

```json
{ "model_name": "deepseek-coder-v2:16b", "base_url": "http://localhost:11434" }
```

Any model available in your local Ollama installation can be used.

## Rules

Rules live in `rules/` (YAML) and `data/custom-rules.json`. The system ships with a set of default rules. You can add project-specific rules via the dashboard → Rules panel.

## License

Apache 2.0
