IntentMesh

OpenAI.Hackathon.Demo.mp4

Slides

The Semantic Linter for the Agent Coding Era

AI writes code fast. Who remembers why?
Your AI conversation stays local. The bugs it prevents don't.

IntentMesh captures developer intents from AI coding conversations and detects when code drifts from those stated requirements.

The Problem

• AI generates 1000s of lines of code per day
• Requirements live in ephemeral chat conversations
• When teammates modify AI-generated code, they don't know the original intent
• Silent regressions happen when code drifts from unstated constraints

The Solution

IntentMesh:

1. Captures intents from AI conversations (Cursor, Claude, Amp)
2. Links them to code via agent-trace.dev attribution
3. Detects drift when code changes violate stated requirements
4. Alerts developers via IDE diagnostics, MCP tools, or CI/CD

Quick Start

VS Code Extension

1. Install the extension (F5 from this repo to run in dev mode)
2. Configure your OpenAI API key in settings (intentmesh.openaiApiKey)
3. Generate code with Cursor/Claude
4. Run node out/cli/write-trace.js . to record attribution
5. Run "IntentMesh: Analyze Changes" to capture intents
6. Make changes → IntentMesh alerts you if they violate intents

MCP Server (Claude Code / Amp)

# Add to your MCP config
codex mcp add intentmesh --env OPENAI_API_KEY=$OPENAI_API_KEY -- node /path/to/intentmesh/out/mcp-stdio.js /path/to/project

Features

• Multi-source intents: AI conversations, PRDs, specs, manual annotations
• Automatic linking: Uses git diffs and agent-trace to map intents to code
• Smart detection: Only flags clear violations (confidence ≥ 0.9)
• Multiple outputs: VS Code diagnostics, MCP server, CLI
• Pluggable architecture: Swap LLM providers, storage backends, intent sources

Architecture

┌─────────────────┐     ┌──────────────────┐     ┌─────────────────┐     ┌──────────────┐
│  AI Conversations│ ──▶ │  Extract Intents │ ──▶ │  Detect Drift   │ ──▶ │  Alert/Block │
│  PRDs & Specs   │     │  (LLM)           │     │  on Changes     │     │  IDE/CI/MCP  │
└─────────────────┘     └──────────────────┘     └─────────────────┘     └──────────────┘

See ARCHITECTURE.md for detailed design.

Configuration

Setting	Description	Default
intentmesh.llmProvider	LLM provider (openai, anthropic, ollama)	openai
intentmesh.openaiApiKey	OpenAI API key	-
intentmesh.llmModel	Model to use	gpt-5-mini

Development

# Install dependencies
npm install

# Build
npm run compile

# Watch mode
npm run watch

# Run extension (press F5 in VS Code)

Project Structure

src/
├── core/           # Interfaces & dependency injection
├── models/         # Data types (Intent, Drift, etc.)
├── sources/        # Intent & attribution sources
├── llm/            # LLM service (LangChain)
├── engine/         # Drift detection logic
├── storage/        # JSON file store
├── adapters/       # MCP server adapter
├── ui/             # VS Code UI (diagnostics, sidebar, hover)
└── extension.ts    # VS Code entry point

Documentation

• Demo Guide - End-to-end demo walkthrough
• Architecture - Detailed system design

License

MIT - see LICENSE

Contributing

Contributions welcome! Please read the architecture doc first to understand the design principles.