In-Context Learning

In-Context Learning is SuperSpec's solution for maintaining context across AI sessions.

The Problem

AI coding assistants don't remember previous sessions:

Session 1: Implemented user authentication
Session 2: AI doesn't know what happened in Session 1
Session 3: Repeat explanations...

This leads to:

• Repeated context setup
• Inconsistent decisions
• Wasted time

The Solution

SuperSpec's sync and resume commands preserve context:

# End of session - save context
superspec sync

# Start of new session - restore context
/ss-resume

How It Works

Sync Command

superspec sync creates context.md containing:

# Context

## Current Change
- Name: userAuth
- Mode: boost
- Status: in_progress

## Git Status
- Modified: src/auth/login.ts
- Added: src/auth/register.ts

## Recent Commits
- feat(auth): add login endpoint
- feat(auth): add user model

## Pending Tasks
- [ ] 4. Add password reset
- [ ] 5. Add email verification

Resume Command

/ss-resume tells the AI to read context.md and continue where you left off.

Usage

Saving Context

Before ending your session:

superspec sync

Or in AI chat:

/ss-sync

Restoring Context

Starting a new session:

/ss-resume

The AI will:

1. Read context.md
2. Understand current state
3. Continue with pending tasks

Best Practices

Sync Frequently

# After significant progress
superspec sync

# Before taking a break
superspec sync

# Before switching to another task
superspec sync

Clean Context

Keep context focused:

# Complete finished changes
superspec archive completedFeature

# Then sync
superspec sync

Verify Resume

After /ss-resume, verify the AI understands:

/ss-resume

> AI: I see we're working on userAuth.
> We've completed tasks 1-3 and are on task 4 (password reset).
> Should I continue with the implementation?

You: Yes, please continue.

Context.md Structure

# Context

## Active Change
Current work focus and status.

## Git Status
Uncommitted changes showing work in progress.

## Recent Activity
What was recently completed.

## Next Steps
What to do next.

Troubleshooting

Context Too Large

If context.md becomes too large:

1. Archive completed changes
2. Commit work in progress
3. Re-run superspec sync

Outdated Context

If context is stale:

# Regenerate context
superspec sync --force

AI Doesn't Understand

If the AI misinterprets context:

The context shows we're working on userAuth,
specifically task 4 about password reset.
Please continue from there.

Next Steps

• Best Practices - Efficiency tips
• CLI Reference - sync command details