[Feature]: Self-Improving Instructions - Auto-suggest CLAUDE.md updates

Summary

Enable Claude to proactively suggest updates to CLAUDE.md and .claude/rules/*.md files based on code changes, user corrections, and repeated questions.

Problem

Current instruction files are static - they become stale as projects evolve:

• User corrections are forgotten between sessions
• New patterns/conventions aren't documented
• Teams repeatedly answer the same questions
• Instructions drift from actual codebase practices

Proposed Solution

Add triggers for Claude to proactively suggest instruction updates:

Trigger 1: Code Changes That Affect Patterns

When Claude introduces new utilities, components, or architectural patterns:

_"I've introduced [new pattern]. Should I update the instructions to document this?"_

Trigger 2: User Correction

When the user corrects Claude for not following a convention:

_"I didn't follow [instruction]. Should I update the rules to make this clearer?"_

Trigger 3: Repeated Questions

When Claude asks about the same topic multiple times:

_"I've asked about [topic] several times. Should I document this?"_

Possible Implementation

• New /learn command to explicitly capture corrections
• Auto-detect patterns that should be documented
• Propose edits to CLAUDE.md with user approval
• Track "instruction violations" to strengthen weak rules

Proof of Concept

We've implemented this pattern manually by adding a "Self-Improving Instructions" section to our CLAUDE.md:

Example: https://github.com/NitinWabale/agents-instructions-collections/blob/main/claude-self-improving-instructions.md

Benefits

| Aspect | Static Instructions | Self-Improving |
|--------|---------------------|----------------|
| Maintenance | Manual updates | Prompted updates |
| Accuracy | Degrades over time | Improves over time |
| Knowledge capture | Often forgotten | Actively captured |
| Team alignment | Inconsistent | Continuously aligned |

Why This Matters

This creates a feedback loop where instructions evolve with the project, making Claude more effective over time without requiring users to manually maintain documentation.