🔮 Weavify - W&B Weave Integration Skill

Add Weights & Biases Weave observability to any LLM project with minimal code changes.

What is Weave?

Weave is W&B's observability toolkit for LLM applications. It automatically traces LLM calls, captures token usage, latency, and costs, and provides a powerful UI for debugging and evaluation.

Key Features

• 🔄 Auto-patching - Automatic tracing for OpenAI, Anthropic, and 20+ providers
• 📊 Token tracking - Full visibility into input/output tokens and costs
• ⏱️ Latency metrics - TTFB and total response time
• 🔍 Debug UI - Visual trace explorer in W&B dashboard
• 📈 Evaluations - Built-in support for LLM evals

---

Installation

Option 1: Vercel Agent Skills (Recommended)

Works with Claude Code, Cursor, Codex, and other AI coding agents:

npx add-skill altryne/weavify-skill

This adds the skill to your project's .cursor/skills/ or agent config.

Option 2: Clawdbot

# Clone to your skills directory
git clone https://github.com/altryne/weavify-skill.git ~/clawd/skills/weave-integration

The skill will be automatically available to your Clawdbot agent.

Option 3: Manual

Copy SKILL.md to your project's agent instructions folder.

---

Quick Start (for your code)

1. Install Weave

# TypeScript/Node.js
npm install weave

# Python
pip install weave

2. Get API Key

Set WANDB_API_KEY environment variable. Get key from wandb.ai/settings.

export WANDB_API_KEY="your-key-here"

3. Initialize & Trace

TypeScript:

import * as weave from 'weave';
import OpenAI from 'openai';

await weave.init('your-team/project-name');

const client = new OpenAI();
// All OpenAI calls are now automatically traced!

Python:

import weave
import openai

weave.init('your-team/project-name')

client = openai.OpenAI()
# All OpenAI calls are now automatically traced!

---

Supported Providers

Weave auto-patches these providers (no code changes needed):

Provider	TypeScript	Python
OpenAI	✅	✅
Anthropic	✅	✅
Google AI	✅	✅
Mistral	✅	✅
Cohere	✅	✅
Groq	✅	✅
Together	✅	✅
LiteLLM	❌	✅
LangChain	❌	✅

---

Manual Tracing

For custom functions, use the @weave.op() decorator:

TypeScript:

const myFunction = weave.op(async (input: string) => {
  // Your code here
  return result;
}, { name: 'myFunction' });

Python:

@weave.op()
def my_function(input: str) -> str:
    # Your code here
    return result

---

Documentation

• 📚 Weave Documentation
• 🚀 Quickstart Guide
• 🔧 TypeScript Integration
• 🐍 Python Integration
• 📊 Tracing Guide
• 📈 Evaluations

---

Skill Structure

weavify-skill/
├── README.md           # This file
├── SKILL.md            # Agent instructions
└── references/
    └── typescript.md   # TypeScript-specific docs

License

MIT

---

Built with 🐺 by Wolfred • Maintained by @altryne