claudify-agent 1.0.1 → 1.0.3

package/.claude/skills/learning-loop/SKILL.md

@@ -1,308 +1,313 @@
-# SKILL.md: Learning Loop
-
-## Description
-Analyzes plan metrics to identify patterns, detect repeated mistakes, and suggest continuous improvements based on historical execution data.
-
-## When to Use
-- Weekly review of plan performance
-- After detecting repeated failures
-- When plan accuracy drops below threshold
-- Proactive optimization of workflows
-
-## Capabilities
-- Pattern recognition from metrics
-- Anti-pattern detection
-- Repeated mistake identification
-- Trend analysis
-- Improvement recommendations
-- Rule update suggestions
-
-## Commands
-/learning-loop-analyze: Full pattern analysis
-/learning-loop-weekly: Weekly performance review
-/learning-loop-suggest: Generate improvement suggestions
-/learning-loop-patterns: Show detected anti-patterns
-
-## Workflows
-
-### Weekly Analysis Workflow
-
-1. **Load Metrics** (30s)
-   ```bash
-   Read: .cursor/plan-metrics.json
-   ```
-   - Last 10-20 plan executions
-   - Extract key metrics
-
-2. **Pattern Detection** (60s)
-   - Run anti-pattern detection rules
-   - Identify repeated mistakes
-   - Calculate trends
-
-3. **Analysis Report** (30s)
-   ```markdown
-   ## Weekly Learning Report
-   
-   ### Performance Summary
-   - Avg Accuracy: X% (trend: ↑/↓/→)
-   - Avg Iterations: X (target: <2)
-   - Tool Efficiency: X% (target: >70%)
-   
-   ### Detected Anti-Patterns
-   1. [SEVERITY] Pattern name
-      - Suggestion: ...
-   
-   ### Repeated Mistakes
-   - Mistake 1 (count: X)
-   - Mistake 2 (count: X)
-   
-   ### Top 3 Improvements
-   1. [Action item]
-   2. [Action item]
-   3. [Action item]
-   
-   ### Rule Update Suggestions
-   - Suggest updating: .cursor/rules/[file]
-   - Reason: [pattern detected]
-   ```
-
-4. **Action Items** (30s)
-   - Prioritize top 3 improvements
-   - Create tasks for rule updates
-   - Schedule follow-up review
-
-### Anti-Pattern Detection
-
-The skill detects these anti-patterns automatically:
-
-1. **Over-Engineering** (High)
-   - Using expensive tools when cheap alternatives work
-   - Detection: avg tool efficiency < 50% over 3 plans
-   - Fix: Follow tool escalation protocol
-
-2. **Planning Without Research** (Critical)
-   - Creating plans without SemanticSearch
-   - Detection: No SemanticSearch in last 3 plans
-   - Fix: Always research before planning
-
-3. **Repeated Execution Failures** (Critical)
-   - Multiple consecutive low-accuracy plans
-   - Detection: 2+ plans with <60% accuracy
-   - Fix: Improve validation process
-
-4. **MCP Overuse** (Medium)
-   - Using MCP tools unnecessarily
-   - Detection: High MCP usage + low accuracy
-   - Fix: Use MCP only when essential
-
-5. **Skipping Validation** (High)
-   - Not running ReadLints/tests
-   - Detection: No validation tools in 3 plans
-   - Fix: Auto-validate after changes
-
-6. **Context Bloat** (Medium)
-   - Loading too much context
-   - Detection: Low efficiency + high iterations
-   - Fix: Apply tiered injection, minimalism
-
-### Trend Analysis
-
-Analyzes performance trends:
-
-```typescript
-if (secondHalfAvg > firstHalfAvg + 5) {
-  trend = "improving";
-  // Reinforce what's working
-} else if (secondHalfAvg < firstHalfAvg - 5) {
-  trend = "declining";
-  // Investigate root causes
-} else {
-  trend = "stable";
-  // Look for optimization opportunities
-}
-```
-
-## Examples
-
-### Example 1: Weekly Review
-
-User: "/learning-loop-weekly"
-
-Agent analyzes last 10 plans:
-
-```markdown
-## Weekly Learning Report (Mar 23-30, 2026)
-
-### Performance Summary
-- Avg Accuracy: 82.5% (↑ +5.2% from last week)
-- Avg Iterations: 1.8 (↓ -0.5, improving)
-- Tool Efficiency: 75.0% (→ stable)
-- Avg Parallel Subagents: 2.3
-
-### Wins This Week
-✅ Accuracy improved above 80% threshold
-✅ Iteration count decreasing
-✅ MCP usage correlated with success
-
-### Detected Anti-Patterns
-1. [MEDIUM] MCP Overuse (2 occurrences)
-   - Suggestion: Use MCP tools only when essential
-
-### Repeated Mistakes
-- None detected this week 🎉
-
-### Top 3 Improvements
-1. Continue current research practices
-2. Reduce MCP usage for simple tasks
-3. Document successful patterns
-
-### Rule Updates
-No updates needed this week
-```
-
-### Example 2: Pattern Detection
-
-User: "/learning-loop-patterns"
-
-Agent detects patterns:
-
-```markdown
-## Detected Anti-Patterns
-
-### 1. [CRITICAL] Planning Without Research
-**Occurrences**: 3 times in last 5 plans
-**Impact**: Avg accuracy 58% (vs 82% when researched)
-**Suggestion**: Always use SemanticSearch before planning
-
-### 2. [HIGH] Skipping Validation
-**Occurrences**: 4 times in last 5 plans
-**Impact**: 75% of unvalidated plans had lints
-**Suggestion**: Enable auto-validate hook
-
-### Recommended Actions
-1. Add rule: "Always research before planning"
-2. Enable auto-validate in hooks.config.json
-3. Create checklist for plan creation
-```
-
-### Example 3: Improvement Suggestions
-
-User: "/learning-loop-suggest"
-
-Agent generates suggestions:
-
-```markdown
-## Improvement Suggestions
-
-### Immediate (This Week)
-1. **Add pre-plan research step**
-   - Time cost: +2 min per plan
-   - Benefit: +20% accuracy improvement
-   
-2. **Enable automatic validation**
-   - Already configured, just use it
-   - Benefit: Catch 100% of lint issues
-
-3. **Reduce MCP tool usage**
-   - Current: 4.2 MCP tools/plan
-   - Target: 2.0 MCP tools/plan
-   - Savings: ~30% token cost
-
-### Medium-term (This Month)
-1. Create validation checklist
-2. Document successful patterns
-3. Update anti-pattern detection rules
-
-### Long-term (This Quarter)
-1. Integrate with CI/CD metrics
-2. A/B test different workflows
-3. Build pattern recognition dashboard
-```
-
-## Best Practices
-
-### 1. Review Weekly
-- Schedule fixed time (e.g., Friday 4pm)
-- Review last 10 plans
-- Identify top 3 improvements
-- Update rules if needed
-
-### 2. Act on Patterns
-- Don't just detect - fix
-- Prioritize critical patterns first
-- Track improvement over time
-
-### 3. Keep Rules Minimal
-- Only add rules for repeated patterns
-- Keep rules under 20 lines
-- Use references over content
-
-### 4. Measure Impact
-- Before/after comparisons
-- Track metric changes
-- Adjust based on data
-
-## Anti-Patterns
-
-### ❌ Ignoring Detected Patterns
-Bad: Detect patterns but don't act  
-Good: Create action items, track fixes
-
-### ❌ Over-Analyzing
-Bad: Analyze every single plan  
-Good: Weekly batch analysis
-
-### ❌ Too Many Rules
-Bad: Add rule for every mistake  
-Good: Only for repeated patterns (3+ occurrences)
-
-### ❌ No Follow-up
-Bad: Generate suggestions, forget them  
-Good: Schedule follow-up review
-
-## Troubleshooting
-
-### Not Enough Data
-**Problem**: Less than 3 plans to analyze
-
-**Solution**: 
-- Wait for more data
-- Run analysis after 10+ plans
-- Use qualitative feedback initially
-
-### False Positive Patterns
-**Problem**: Pattern detected but not actually an issue
-
-**Solution**:
-- Adjust detection thresholds
-- Add exception rules
-- Manual review before acting
-
-### No Patterns Detected
-**Problem**: Analysis shows no patterns
-
-**Solution**:
-- Check if thresholds too strict
-- Look for subtle patterns
-- Consider qualitative feedback
-
-## Metrics Reference
-
-### Key Metrics Tracked
-- `plan_accuracy`: % of plan followed correctly
-- `iteration_count`: Number of /grind loops
-- `tool_efficiency`: Cost-weighted tool score
-- `parallel_subagents`: Number of parallel explorations
-- `mcp_usage`: MCP server utilization
-- `cost_estimate`: Estimated token consumption
-
-### Thresholds
-- **Excellent**: Accuracy >90%, Iterations <2, Efficiency >80%
-- **Good**: Accuracy 70-90%, Iterations 2-3, Efficiency 60-80%
-- **Needs Work**: Accuracy <70%, Iterations >3, Efficiency <60%
-
-## Related Documentation
-- `AGENT-INSTRUCTION-BEST-PRACTICES.md` - Core research
-- `.cursor/hooks.config.json` - Configuration
-- `.cursor/plan-metrics.json` - Historical data
-- `.cursor/docs/PLAN-MODE-OPTIMIZATION.md` - Plan mode guide
+---
+name: learning-loop
+description: Analyze plan metrics in .claude/plan-metrics.json to detect anti-patterns and repeated mistakes and suggest rule improvements. Use for weekly reviews or when plan accuracy drops.
+---
+
+# SKILL.md: Learning Loop
+
+## Description
+Analyzes plan metrics to identify patterns, detect repeated mistakes, and suggest continuous improvements based on historical execution data.
+
+## When to Use
+- Weekly review of plan performance
+- After detecting repeated failures
+- When plan accuracy drops below threshold
+- Proactive optimization of workflows
+
+## Capabilities
+- Pattern recognition from metrics
+- Anti-pattern detection
+- Repeated mistake identification
+- Trend analysis
+- Improvement recommendations
+- Rule update suggestions
+
+## Commands
+/learning-loop-analyze: Full pattern analysis
+/learning-loop-weekly: Weekly performance review
+/learning-loop-suggest: Generate improvement suggestions
+/learning-loop-patterns: Show detected anti-patterns
+
+## Workflows
+
+### Weekly Analysis Workflow
+
+1. **Load Metrics** (30s)
+   ```bash
+   Read: .claude/plan-metrics.json
+   ```
+   - Last 10-20 plan executions
+   - Extract key metrics
+
+2. **Pattern Detection** (60s)
+   - Run anti-pattern detection rules
+   - Identify repeated mistakes
+   - Calculate trends
+
+3. **Analysis Report** (30s)
+   ```markdown
+   ## Weekly Learning Report
+   
+   ### Performance Summary
+   - Avg Accuracy: X% (trend: ↑/↓/→)
+   - Avg Iterations: X (target: <2)
+   - Tool Efficiency: X% (target: >70%)
+   
+   ### Detected Anti-Patterns
+   1. [SEVERITY] Pattern name
+      - Suggestion: ...
+   
+   ### Repeated Mistakes
+   - Mistake 1 (count: X)
+   - Mistake 2 (count: X)
+   
+   ### Top 3 Improvements
+   1. [Action item]
+   2. [Action item]
+   3. [Action item]
+   
+   ### Rule Update Suggestions
+   - Suggest updating: .claude/rules/[file]
+   - Reason: [pattern detected]
+   ```
+
+4. **Action Items** (30s)
+   - Prioritize top 3 improvements
+   - Create tasks for rule updates
+   - Schedule follow-up review
+
+### Anti-Pattern Detection
+
+The skill detects these anti-patterns automatically:
+
+1. **Over-Engineering** (High)
+   - Using expensive tools when cheap alternatives work
+   - Detection: avg tool efficiency < 50% over 3 plans
+   - Fix: Follow tool escalation protocol
+
+2. **Planning Without Research** (Critical)
+   - Creating plans without Explore (quick)
+   - Detection: No Explore (quick) in last 3 plans
+   - Fix: Always research before planning
+
+3. **Repeated Execution Failures** (Critical)
+   - Multiple consecutive low-accuracy plans
+   - Detection: 2+ plans with <60% accuracy
+   - Fix: Improve validation process
+
+4. **MCP Overuse** (Medium)
+   - Using MCP tools unnecessarily
+   - Detection: High MCP usage + low accuracy
+   - Fix: Use MCP only when essential
+
+5. **Skipping Validation** (High)
+   - Not running Lint check/tests
+   - Detection: No validation tools in 3 plans
+   - Fix: Auto-validate after changes
+
+6. **Context Bloat** (Medium)
+   - Loading too much context
+   - Detection: Low efficiency + high iterations
+   - Fix: Apply tiered injection, minimalism
+
+### Trend Analysis
+
+Analyzes performance trends:
+
+```typescript
+if (secondHalfAvg > firstHalfAvg + 5) {
+  trend = "improving";
+  // Reinforce what's working
+} else if (secondHalfAvg < firstHalfAvg - 5) {
+  trend = "declining";
+  // Investigate root causes
+} else {
+  trend = "stable";
+  // Look for optimization opportunities
+}
+```
+
+## Examples
+
+### Example 1: Weekly Review
+
+User: "/learning-loop-weekly"
+
+Agent analyzes last 10 plans:
+
+```markdown
+## Weekly Learning Report (Mar 23-30, 2026)
+
+### Performance Summary
+- Avg Accuracy: 82.5% (↑ +5.2% from last week)
+- Avg Iterations: 1.8 (↓ -0.5, improving)
+- Tool Efficiency: 75.0% (→ stable)
+- Avg Parallel Subagents: 2.3
+
+### Wins This Week
+✅ Accuracy improved above 80% threshold
+✅ Iteration count decreasing
+✅ MCP usage correlated with success
+
+### Detected Anti-Patterns
+1. [MEDIUM] MCP Overuse (2 occurrences)
+   - Suggestion: Use MCP tools only when essential
+
+### Repeated Mistakes
+- None detected this week 🎉
+
+### Top 3 Improvements
+1. Continue current research practices
+2. Reduce MCP usage for simple tasks
+3. Document successful patterns
+
+### Rule Updates
+No updates needed this week
+```
+
+### Example 2: Pattern Detection
+
+User: "/learning-loop-patterns"
+
+Agent detects patterns:
+
+```markdown
+## Detected Anti-Patterns
+
+### 1. [CRITICAL] Planning Without Research
+**Occurrences**: 3 times in last 5 plans
+**Impact**: Avg accuracy 58% (vs 82% when researched)
+**Suggestion**: Always use Explore (quick) before planning
+
+### 2. [HIGH] Skipping Validation
+**Occurrences**: 4 times in last 5 plans
+**Impact**: 75% of unvalidated plans had lints
+**Suggestion**: Enable auto-validate hook
+
+### Recommended Actions
+1. Add rule: "Always research before planning"
+2. Enable auto-validate in hooks.config.json
+3. Create checklist for plan creation
+```
+
+### Example 3: Improvement Suggestions
+
+User: "/learning-loop-suggest"
+
+Agent generates suggestions:
+
+```markdown
+## Improvement Suggestions
+
+### Immediate (This Week)
+1. **Add pre-plan research step**
+   - Time cost: +2 min per plan
+   - Benefit: +20% accuracy improvement
+   
+2. **Enable automatic validation**
+   - Already configured, just use it
+   - Benefit: Catch 100% of lint issues
+
+3. **Reduce MCP tool usage**
+   - Current: 4.2 MCP tools/plan
+   - Target: 2.0 MCP tools/plan
+   - Savings: ~30% token cost
+
+### Medium-term (This Month)
+1. Create validation checklist
+2. Document successful patterns
+3. Update anti-pattern detection rules
+
+### Long-term (This Quarter)
+1. Integrate with CI/CD metrics
+2. A/B test different workflows
+3. Build pattern recognition dashboard
+```
+
+## Best Practices
+
+### 1. Review Weekly
+- Schedule fixed time (e.g., Friday 4pm)
+- Review last 10 plans
+- Identify top 3 improvements
+- Update rules if needed
+
+### 2. Act on Patterns
+- Don't just detect - fix
+- Prioritize critical patterns first
+- Track improvement over time
+
+### 3. Keep Rules Minimal
+- Only add rules for repeated patterns
+- Keep rules under 20 lines
+- Use references over content
+
+### 4. Measure Impact
+- Before/after comparisons
+- Track metric changes
+- Adjust based on data
+
+## Anti-Patterns
+
+### ❌ Ignoring Detected Patterns
+Bad: Detect patterns but don't act  
+Good: Create action items, track fixes
+
+### ❌ Over-Analyzing
+Bad: Analyze every single plan  
+Good: Weekly batch analysis
+
+### ❌ Too Many Rules
+Bad: Add rule for every mistake  
+Good: Only for repeated patterns (3+ occurrences)
+
+### ❌ No Follow-up
+Bad: Generate suggestions, forget them  
+Good: Schedule follow-up review
+
+## Troubleshooting
+
+### Not Enough Data
+**Problem**: Less than 3 plans to analyze
+
+**Solution**: 
+- Wait for more data
+- Run analysis after 10+ plans
+- Use qualitative feedback initially
+
+### False Positive Patterns
+**Problem**: Pattern detected but not actually an issue
+
+**Solution**:
+- Adjust detection thresholds
+- Add exception rules
+- Manual review before acting
+
+### No Patterns Detected
+**Problem**: Analysis shows no patterns
+
+**Solution**:
+- Check if thresholds too strict
+- Look for subtle patterns
+- Consider qualitative feedback
+
+## Metrics Reference
+
+### Key Metrics Tracked
+- `plan_accuracy`: % of plan followed correctly
+- `iteration_count`: Number of fix-and-retest loops
+- `tool_efficiency`: Cost-weighted tool score
+- `parallel_subagents`: Number of parallel explorations
+- `mcp_usage`: MCP server utilization
+- `cost_estimate`: Estimated token consumption
+
+### Thresholds
+- **Excellent**: Accuracy >90%, Iterations <2, Efficiency >80%
+- **Good**: Accuracy 70-90%, Iterations 2-3, Efficiency 60-80%
+- **Needs Work**: Accuracy <70%, Iterations >3, Efficiency <60%
+
+## Related Documentation
+- `AGENT-INSTRUCTION-BEST-PRACTICES.md` - Core research
+- `.claude/hooks.config.json` - Configuration
+- `.claude/plan-metrics.json` - Historical data
+- `.claude/docs/PLAN-MODE-OPTIMIZATION.md` - Plan mode guide