Multi-AI Support

The AI Context System works with any AI coding assistant. The core is universal markdown files—no tool-specific features required.

Universal by Design

The system is built on three markdown files:

File	Purpose
CLAUDE.md	Entry point, project overview
context/STATUS.md	Current state, Working Set, next actions
context/DECISIONS.md	Append-only decision log

Any AI that can read and write files can use this system.

How It Works with Different Tools

Claude Code (Slash Commands)

Claude Code gets convenience commands:

/init-context  # Creates the three context files
/save          # Updates STATUS.md at session end

But these commands just automate what you could do manually—read and write markdown files.

Any Other AI Tool

Works identically, just without slash commands:

1. Read CLAUDE.md to understand the project
2. Read context/STATUS.md for current state
3. Work on files in the Working Set
4. Update STATUS.md when done
5. Append to DECISIONS.md for significant decisions

The Session Loop is the same:

Start → Read STATUS.md
Work → Edit Working Set files
End → Update STATUS.md

Examples by Tool

Cursor

1. Open project
2. @ context/STATUS.md to see current state
3. Work on files
4. Ask Cursor to update STATUS.md:
   "Update context/STATUS.md with:
    - Objective: [current goal]
    - Working Set: [files you touched]
    - Next Actions: [what's next]"

Aider

# See current state
aider --read context/STATUS.md

# Work on files
aider src/feature.ts

# Update status
aider context/STATUS.md
# Ask to update fields

GitHub Copilot / Any AI

1. Open context/STATUS.md
2. Read current state
3. Make your changes
4. Update STATUS.md manually or ask the AI

API-Based AIs

// Read context before generating code
const status = await readFile('context/STATUS.md');
const decisions = await readFile('context/DECISIONS.md');

// Include in prompt
const prompt = `
Project context:
${status}

Past decisions:
${decisions}

Task: ${userTask}
`;

The File Structure

your-project/
├── CLAUDE.md               # Entry point (auto-loaded by Claude Code)
├── context/
│   ├── STATUS.md           # Current state
│   └── DECISIONS.md        # Decision history
└── .claude/
    └── commands/           # Claude Code slash commands (optional)

Every AI tool can:

• Read these markdown files
• Understand the structure
• Update documentation
• Reference past decisions

AI-to-AI Handoffs

The system enables seamless handoffs between different AI tools:

Example: Claude Code → Cursor

Day 1: Claude Code

/save

STATUS.md now contains:

Objective: Implement user authentication
Working Set: src/auth/login.ts, src/auth/session.ts
Next Actions: Add password hashing, implement JWT tokens

Day 2: Cursor

@ context/STATUS.md
"Continue from where the last session left off"

Cursor reads the same STATUS.md and knows exactly what to do.

Why It Works

• Same files — Both AIs read/write the same markdown
• Same format — STATUS.md structure is AI-agnostic
• Same information — Objective, Working Set, Next Actions
• Zero context loss — Everything persists in files

Claude Code: Why Optimized?

Claude Code gets extra conveniences:

Feature	Claude Code	Other AIs
Slash commands	/save, /init-context	Manual updates
Auto-load	CLAUDE.md loads automatically	Must read manually
Command discovery	Type / to see all commands	N/A

But the core system is identical. Claude Code commands just automate:

• Reading files
• Updating fields
• Formatting entries

You can do all of this manually in any tool.

Best Practices

Keep Context Updated

After significant work, update STATUS.md:

LastUpdated: [today]
HeadCommit: [current git SHA]
Objective: [what you're working toward]

## Working Set
- [files you're touching]

## Next Actions
- [what's next]

## Blocked On
- [blockers, or "(None)"]

Document Decisions

When making choices with tradeoffs:

## YYYY-MM-DD: [Area] Decision Title
Why: [reason]
Tradeoff: [what you gave up]
RevisitWhen: [trigger to revisit]

Document Tool Switches

When switching between AI tools:

## Next Actions
- Continue auth implementation (switching from Claude Code to Cursor)
- See DECISIONS.md for JWT approach rationale

Success Metric

"I can switch between Claude Code, Cursor, Aider, or any other AI tool and continue exactly where I left off."

When this is true, multi-AI support is working.

Next Steps

• Getting Started — Installation
• Session Continuity — How context persists
• STATUS.md Guide — Current state format
• DECISIONS.md Guide — Decision format