v1.3.0 - Meta-Learning (Reflect-on-Reflect)

What's New

Meta-Learning System

The reflect system now learns from your own feedback to improve pattern detection over time.

New Features

• /reflect-meta command: View pattern performance statistics
• --use-meta flag: Enable score-based confidence adjustments
• Passive feedback logging: Decisions (Accept/Modify/Skip) are automatically recorded
• Pattern health tracking: See which patterns work well and which need review

Design Principles

• Non-invasive: Meta-learning is completely passive by default
• Opt-in activation: Score-based adjustments require explicit --use-meta flag
• Non-blocking: Meta-learning failures never break the core reflect workflow
• Transparent: All data stored in human-readable JSON locally

New Files

File	Purpose
scripts/meta_learning.py	Core meta-learning module
commands/reflect-meta.md	Command documentation

Thresholds

Status	Acceptance Rate	Adjustment
Excellent	>80%	+0.1 boost
Healthy	50-80%	No change
Needs Review	20-50%	-0.15 penalty
Deprecated	<20%	-0.3 penalty, flagged

Storage

• ~/.claude/reflect/meta/feedback-log.jsonl - Decision history
• ~/.claude/reflect/meta/pattern-scores.json - Computed scores

---

Full Changelog: v1.2.0...v1.3.0