Self-Improve Skill

Type: Meta Trigger: /learn:apply, after analysis Auto-invoke: No (requires user approval)

---

Purpose

Apply learned improvements to the Aura Frog plugin:

• Update rules based on patterns
• Adjust agent routing
• Modify workflow configurations
• Generate new knowledge base entries

---

Usage

/learn:apply                    # Review and apply pending improvements
/learn:apply --auto             # Auto-apply high-confidence (≥0.8) suggestions
/learn:apply --preview          # Preview changes without applying
/learn:apply --id <pattern_id>  # Apply specific pattern

---

Improvement Types

1. Rule Updates

When patterns suggest rule modifications:

# Before analysis
rules:
  code_coverage_threshold: 80

# After learning shows 85% coverage correlates with fewer bugs
rules:
  code_coverage_threshold: 85

Implementation:

1. Identify pattern suggesting rule change
2. Generate rule modification
3. Create backup of current rule
4. Apply modification
5. Mark pattern as applied in Supabase

2. Agent Routing

When agent success rates indicate better routing:

# Update agent-detector scoring
agent_overrides:
  react_component:
    primary: react-expert
    confidence_boost: 10
  api_endpoint:
    primary: nodejs-expert
    secondary: python-expert

Implementation:

1. Query agent success rates by task type
2. Calculate optimal routing
3. Update agent-detector configuration
4. Test with sample tasks

3. Workflow Adjustments

When workflow patterns suggest changes:

# Add auto-stop threshold adjustment
workflow:
  phase_2:
    timeout: 300  # increased from 180
    batch_size: 50  # for large test suites

4. Knowledge Base

When insights should be preserved:

# New knowledge entry
Type: tip
Title: Use TDD for API endpoints
Content: Analysis of 47 workflows shows TDD approach for API
         endpoints reduces bug count by 40% compared to non-TDD.
Tags: [workflow, tdd, api]
Priority: 75

---

Safety Guards

Approval Required

All improvements require explicit approval unless:

• --auto flag is used AND
• Confidence ≥ 0.8 AND
• Pattern frequency ≥ 5

Rollback Support

Every applied improvement:

1. Creates backup of original configuration
2. Logs the change with timestamp
3. Can be rolled back via /learn:rollback <change_id>

Validation

Before applying:

1. Syntax validation of generated rules
2. Conflict detection with existing rules
3. Impact assessment (files affected)

---

Apply Process

Step 1: Fetch Pending Improvements

SELECT * FROM v_improvement_suggestions
WHERE applied = FALSE
ORDER BY confidence DESC;

Step 2: Generate Changes

For each suggestion:

1. Determine target file(s)
2. Generate modification
3. Calculate impact

Step 3: Present for Review

## Pending Improvements

### 1. Increase Phase 2 timeout (ID: abc123)
**Confidence:** 85%
**Frequency:** 8 occurrences
**Impact:** Modifies `ccpm-config.yaml`

**Change:**
```diff
workflow:
  phase_5a:
-   timeout: 180
+   timeout: 300

Evidence:

• 8 workflows failed with timeout in Phase 2
• Average test count: 127 files
• Average completion time needed: 245s

[ ] Apply [ ] Skip [ ] Modify

### Step 4: Apply Changes

1. Create backup: `backups/ccpm-config.yaml.20260107`
2. Apply modification
3. Update Supabase: `UPDATE af_learned_patterns SET applied=TRUE WHERE id=...`
4. Log change

---

## Rollback

```bash
/learn:rollback <change_id>     # Rollback specific change
/learn:rollback --list          # List recent changes
/learn:rollback --all           # Rollback all changes from today

---

Example Session

User: /learn:apply

Claude: 🐸 Self-Improve: Found 3 pending improvements

## Improvement 1/3: Agent Routing Update
Confidence: 92% | Frequency: 28

**Suggestion:** Default react-expert for .tsx files

**Evidence:**
- 28 workflows with React component tasks
- react-expert success rate: 96%
- Current default (ui-expert) success rate: 78%

**Change to:** `agents/smart-agent-detector.md`
```diff
react_patterns:
- default_agent: ui-expert
+ default_agent: react-expert
+ file_extensions: [.tsx, .jsx]

Apply this improvement? [y/n/skip/modify]:

User: y

Claude: ✅ Applied improvement 1/3 Backup saved to: backups/smart-agent-detector.md.20260107

Improvement 2/3: ...

---

## Configuration

In `ccpm-config.yaml`:

```yaml
learning:
  self_improve:
    enabled: true
    auto_apply_threshold: 0.8  # Confidence threshold for --auto
    min_frequency: 5           # Minimum occurrences for auto-apply
    backup_dir: backups/       # Where to store backups
    max_auto_per_day: 10       # Limit auto-applies per day

---

Version: 1.0.0