Context Saver

Context Saver Skill

This skill enables Claude Capsule Kit to save discoveries to persistent storage.

Purpose

Problem: Claude sessions are stateless - discoveries are lost when session ends. Solution: Save findings to docs/exploration/ files that next Claude reads.

When to Use

Use this skill when:

• Completing an exploration phase
• Making important architectural discoveries
• Finding bugs or issues
• Learning how a system works
• Finishing a significant task

How It Works

1. Determine what to save: Summarize key findings from current work
2. Choose appropriate file:
   • CURRENT_SESSION.md - Update with latest progress
   • PHASE_X_FINDINGS.md - Save phase-specific discoveries
   • DISCOVERY_[TOPIC].md - Document specific insights
3. Write structured markdown: Clear, scannable format for future Claude
4. Update session state: Track what's done, what's next

File Structure

CURRENT_SESSION.md

# Current Session: [Goal]
**Date**: YYYY-MM-DD
**Status**: In Progress / Complete

## Completed
- [x] Task 1
- [x] Task 2

## Discovered
- Finding 1 with file references
- Finding 2 with code examples

## Next Steps
- [ ] Continue with...

PHASE_X_FINDINGS.md

# Phase X: [Phase Name]
**Date**: YYYY-MM-DD
**Status**: Complete

## Summary
3-5 sentence overview of what was explored.

## Key Findings
1. **Finding 1**: Details with file paths
2. **Finding 2**: Details with code snippets

## Files Explored
- `/path/to/file:123` - What we learned

Example Usage

Situation: Just completed exploring Labs database schema

Skill Execution:

1. Create /docs/exploration/LABS_DATABASE_SCHEMA.md
2. Document findings:
   • 3 main tables (experiments, experiment_runs, validation_rules)
   • Foreign key relationships
   • JSONB config storage patterns
   • Confidence score calculation (DECIMAL 3,2)
3. Update CURRENT_SESSION.md with progress
4. Mark task as completed

Result: Next Claude session reads LABS_DATABASE_SCHEMA.md and knows exactly what was discovered.

Best Practices

DO:

• ✅ Save concrete findings (file paths, line numbers, code snippets)
• ✅ Use clear section headers for scannability
• ✅ Include "Next Steps" for continuity
• ✅ Reference specific files explored
• ✅ Document both what works AND what doesn't

DON'T:

• ❌ Save vague statements ("looked at some code")
• ❌ Omit file references
• ❌ Write walls of text without structure
• ❌ Forget to update CURRENT_SESSION.md

Integration with SessionStart Hook

When next Claude session starts:

1. SessionStart hook detects files in docs/exploration/
2. Prints: "🧠 Claude Capsule Kit MEMORY LOADED: Previous exploration findings available"
3. Lists all saved files
4. Next Claude reads relevant files to continue work

This creates persistent memory across sessions!

Commands

# List saved findings
ls docs/exploration/

# Read a specific finding
cat docs/exploration/PHASE_1_FINDINGS.md

# Update current session
# (Edit CURRENT_SESSION.md with new progress)

Success Criteria

✅ Findings are clear enough that another developer could understand ✅ File paths included for easy navigation ✅ Next steps documented for continuity ✅ CURRENT_SESSION.md always reflects latest state

Remember: This skill is THE KEY to making Claude truly "SUPER" - it's how we remember and build on past work!