using-live-documentation

# Using Live Documentation

## Overview

**Your training data is outdated. Current documentation is always more accurate.**

When implementing features, answering questions, or debugging issues involving libraries/frameworks/tools, you MUST fetch current documentation using Context7 before writing code or making recommendations.

## Core Principle

LLM training data becomes stale the moment training ends. Libraries evolve:
- APIs change between versions
- Best practices get updated
- New features get added
- Old patterns get deprecated

**Never implement from memory. Always verify with current docs.**

## Mandatory Workflow

### Step 1: Recognize the Trigger

You MUST use documentation search when you encounter ANY of these:

- Library name mentioned (react-query, fastapi, pydantic, express, etc.)
- Framework name mentioned (Next.js, Django, React, Vue, etc.)
- Version number specified (react-query v5, Python 3.12, etc.)
- Technical concept tied to specific tool (optimistic updates in react-query)
- Implementation questions (how do I X in Y?)
- Best practices questions (what's the right way to X?)
- Debugging library-specific behavior

**Red flags that mean you're about to fail:**
- "Based on my knowledge of..."
- "From what I remember about..."
- "The typical pattern for..."
- Writing code without checking docs first
- Having uncertainty about the correct approach

### Step 2: Dispatch Documentation Search Subagent

**Why subagent instead of direct Context7:**
- Saves 10,000-20,000 tokens of context in main agent
- Subagent filters docs to only what you need
- Main agent stays focused on implementation
- Better token management across the session

**How to dispatch:**

Dispatch the documentation-searcher agent with the following information:

- **Library name**: Exact package/library name (e.g., "react-query", "fastapi", "pydantic")
- **Topic**: Specific concept or feature (e.g., "optimistic updates", "path parameters", "field validators")
- **What you need**: Spec