claude-plugins/plugins/learn

Claude Learn

A Claude Code plugin for capturing lessons learned during development sessions with domain-based memory, selective saving, and memory hygiene.

Installation

claude plugin install learn@romfast-plugins

Commands

Command	Description
/learn:analyze	Analyze conversation and propose lessons (selective)
/learn:status	Show memory statistics by domain
/learn:import <file>	Import from another memory file
/learn:reflect	Consolidate and clean up memory
/learn:cleanup	Review and remove low-value entries

Features

Domain-based Memory

• Organized by backend, frontend, database, testing, etc.
• Selective loading via paths frontmatter
• Automatic domain detection from files in conversation

Selective Saving (v2.1.0)

The plugin now uses intelligent signal detection to save only valuable lessons:

What Gets Saved (High Priority):

• Explicit commands: "ține minte", "remember this", "memorează"
• Stated preferences: "întotdeauna fă X", "always do Y"
• Repeated corrections: Same issue corrected 2+ times
• Historical references: "cum am discutat", "as we discussed"

What Gets Filtered Out:

• One-time typo fixes
• Single-file specific bugs
• Import path corrections
• Hardcoded value adjustments

Memory Hygiene (v2.1.0)

Each entry now tracks metadata for cleanup:

Field	Description
source	How it was captured: explicit, inferred, or repeated
confidence	Signal strength: high, medium, or low
reinforced	Times used in subsequent sessions
trigger	Original phrase that triggered saving

Run /learn:cleanup periodically to review and remove low-value entries.

SessionEnd Hook

When you exit a session (/exit, /clear, Ctrl+D), the plugin shows a reminder:

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
💡 Tip: Rulează /learn:analyze pentru a captura lecțiile din sesiune
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Note: Requires Claude Code 2.0.50+ for /clear to trigger the hook.

Plugin Structure

learn/
├── .claude-plugin/plugin.json
├── commands/
│   ├── analyze.md     # Main analysis with signal detection
│   ├── cleanup.md     # Memory hygiene command
│   ├── import.md      # Import from other formats
│   ├── reflect.md     # Cross-domain consolidation
│   └── status.md      # Statistics display
├── hooks/hooks.json   # SessionEnd hook config
├── scripts/
│   └── session-end-reminder.sh  # Reminder script
├── rules/memory-format.md
├── templates/
│   ├── domain-memory.md
│   ├── domains-config.md
│   └── memory-entry.json
└── README.md

Memory Format

Entries are saved with metadata:

### API Response Format Standard
**Discovered**: 2025-01-06 (feature: user-api)
**Source**: explicit | **Confidence**: high | **Reinforced**: 3
**Trigger**: "ține minte să folosești acest format"
**Description**: All API responses must use {data, error, meta} structure

**Example** (`src/api/users.ts:45`):
```typescript
return { data: users, error: null, meta: { count: 10 } }

Tags: api, response, format

## Development

```bash
# Test the plugin
claude --plugin-dir ./plugins/learn

# Verify commands
/learn:status
/learn:analyze
/learn:cleanup --dry-run

Changelog

v2.1.0

• Added selective saving with signal detection
• Added /learn:cleanup command for memory hygiene
• Added metadata tracking (source, confidence, reinforced, trigger)
• Improved Stop hook with specific criteria
• Backward compatible with v2.0.0 entries

v2.0.0

• Domain-based memory organization
• Selective loading via paths frontmatter
• Automatic domain detection

License

MIT