claude-reflect

1# Claude Reflect - Self-Learning System
2 
3A two-stage system that helps Claude Code learn from user corrections.
4 
5## How It Works
6 
7**Stage 1: Capture (Automatic)**
8Hooks detect correction patterns ("no, use X", "actually...", "use X not Y") and queue them to `~/.claude/learnings-queue.json`.
9 
10**Stage 2: Process (Manual)**
11User runs `/reflect` to review and apply queued learnings to CLAUDE.md files.
12 
13## Available Commands
14 
15| Command | Purpose |
16|---------|---------|
17| `/reflect` | Process queued learnings with human review |
18| `/reflect --scan-history` | Scan past sessions for missed learnings |
19| `/reflect --dry-run` | Preview changes without applying |
20| `/skip-reflect` | Discard all queued learnings |
21| `/view-queue` | View pending learnings without processing |
22 
23## When to Remind Users
24 
25Remind users about `/reflect` when:
26- They complete a feature or meaningful work unit
27- They make corrections you should remember for future sessions
28- They explicitly say "remember this" or similar
29- Context is about to compact and queue has items
30 
31## Correction Detection Patterns
32 
33High-confidence corrections:
34- Tool rejections (user stops an action with guidance)
35- "no, use X" / "don't use Y"
36- "actually..." / "I meant..."
37- "use X not Y" / "X instead of Y"
38- "remember:" (explicit marker)
39 
40## CLAUDE.md Destinations
41 
42- `~/.claude/CLAUDE.md` - Global learnings (model names, general patterns)
43- `./CLAUDE.md` - Project-specific learnings (conventions, tools, structure)
44 
45## Example Interaction
46…