CONTEXT.md Toolkit - Claude Code Plugin

A comprehensive Claude Code plugin that implements the CONTEXT.md workflow system for managing human-authored context and agent-ready briefings.

Overview

This plugin bridges the gap between how humans naturally express ideas and how AI agents optimally process information. Instead of forcing humans to write in "robot speak," it provides a workflow where you can:

1. Capture ideas naturally in CONTEXT.md (human-friendly scratchpad)
2. Convert to structured briefings in CLAUDE.md (agent-optimized format)
3. Keep both synchronized as your project evolves

Core Concept

The CONTEXT.md system recognizes that:

• Humans express ideas best in conversational, narrative form
• AI agents work best with structured, concise briefings
• LLMs excel at converting between these two "dialects"
• A single markdown file is often sufficient for project context (vs. complex RAG systems)

Two complementary files:

• CONTEXT.md: Expressive, detailed, human-friendly context (often from voice transcription)
• CLAUDE.md: Structured, concise, agent-ready briefing (automatically read by Claude Code)

Installation

From GitHub

/plugin install danielrosehill/Claude-Code-Context-Toolkit

From Local Directory

/plugin install /path/to/Claude-Code-Context-Toolkit

Features

Slash Commands

Core Operations

/context-to-claude - Convert CONTEXT.md to CLAUDE.md

• Proofreads and lightly edits CONTEXT.md (fixes typos, adds punctuation)
• Extracts key requirements and creates structured CLAUDE.md
• Preserves human voice in CONTEXT.md while creating agent-optimized briefing

/claude-to-context - Extract info from CLAUDE.md to CONTEXT.md

• Identifies valuable information in CLAUDE.md
• Incorporates it into CONTEXT.md naturally
• Maintains narrative, conversational style

Context Management

/add-to-context - Add new information to context pool

• Integrates new context into CONTEXT.md or context-data/
• Updates CLAUDE.md to reflect new information
• Maintains organization and avoids duplication

/remove-from-context - Remove outdated information

• Removes specified content from context pool
• Updates CLAUDE.md for consistency
• Cleans up document structure

/manual-context-update - Guided manual editing workflow

• Interactive workflow for coordinated updates
• Supports updates to either or both files
• Validates consistency after changes

/chunk-repo-context - Split CONTEXT.md into organized chunks

• Analyzes content and proposes topic-based organization
• Creates context-data/ directory with focused files
• Builds index and updates CLAUDE.md references

Specialized Agents

context-converter - Handles conversion between context formats

• Optimized for CONTEXT.md ↔ CLAUDE.md conversion
• Preserves distinct character of each file type
• Maintains consistency and quality

context-manager - Manages context file organization

• Adds, removes, and reorganizes context
• Handles chunking and restructuring
• Maintains cross-references and indexes

Hooks

context-file-detector (on_repo_open)

• Detects CONTEXT.md or context-data/ in repository
• Reminds agent to read context files
• Runs automatically when opening repository

context-sync-reminder (on_file_write)

• Triggers when context files are modified
• Reminds to keep CONTEXT.md and CLAUDE.md synchronized
• Suggests relevant slash commands

Workflow Examples

Starting a New Project

1. Create CONTEXT.md and describe your project naturally (use voice transcription if you like)
2. Run /context-to-claude to generate CLAUDE.md
3. Continue working with Claude Code - it now has optimal context

Adding New Requirements

1. Add information to CONTEXT.md (conversationally)
2. Run /add-to-context to integrate cleanly
3. CLAUDE.md is automatically updated

Growing Project Context

When CONTEXT.md gets large:

1. Run /chunk-repo-context
2. Review proposed organization
3. Approve to split into topic-based files in context-data/

File Structure

your-repo/
├── CONTEXT.md              # Human-friendly context scratchpad
├── CLAUDE.md              # Agent-ready briefing (auto-read by Claude Code)
└── context-data/          # Optional: chunked context (for large projects)
    ├── README.md          # Index of context files
    ├── project-vision.md
    ├── requirements.md
    └── technical-specs.md

Best Practices

CONTEXT.md

• Write naturally and expressively
• Use voice transcription if it helps you capture ideas
• Don't worry about being too verbose
• Include the "why" behind decisions
• Preserve your thought process

CLAUDE.md

• Keep it structured and actionable
• Extract specific requirements
• Provide clear implementation guidance
• Reference CONTEXT.md for deeper context
• Update when requirements change

Synchronization

• Run /context-to-claude when CONTEXT.md changes significantly
• Use /add-to-context to keep both files aligned
• Let hooks remind you when files get out of sync

Chunking

• Chunk when CONTEXT.md exceeds ~500-1000 lines
• Organize by topic, not arbitrary size
• Use descriptive filenames (kebab-case)
• Maintain index in context-data/README.md

Related Projects

• CONTEXT.md Specification - Original concept and notes
• Context Cruncher - CLI tools for context processing

Author

Daniel Rosehill

• Website: danielrosehill.com
• Business: DSR Holdings
• Email: [email protected]

License

MIT License - See LICENSE file for details

Feedback

Issues, suggestions, and feature requests: GitHub Issues