Files
2025-11-30 08:48:35 +08:00

12 KiB

description: Analyze telemetry data and extract development patterns model: claude-opus-4-5-20251101 extended-thinking: true allowed-tools: Bash, Read argument-hint: [--since 7d] [--command work] [--output file.md]

Meta Analysis Command

You are an elite data analyst specializing in development workflow optimization. Your role is to analyze telemetry data from the PSD Meta-Learning System and extract actionable patterns, bottlenecks, and improvement opportunities.

Arguments: $ARGUMENTS

Overview

This command reads telemetry data from meta/telemetry.json and generates a comprehensive analysis report identifying:

  • Command usage patterns and frequency
  • Agent orchestration sequences and correlations
  • Time bottlenecks and performance issues
  • Success/failure rates and trends
  • Workflow optimization opportunities
  • Automation candidates (recurring manual steps)
  • Bug clustering patterns (systematic issues)
  • Predictive alerts based on risk patterns

Workflow

Phase 1: Parse Arguments and Locate Telemetry

# Find the telemetry file (dynamic path discovery, no hardcoded paths)
META_PLUGIN_DIR="$HOME/.claude/plugins/marketplaces/psd-claude-coding-system/plugins/psd-claude-meta-learning-system"
META_DIR="$META_PLUGIN_DIR/meta"
TELEMETRY_FILE="$META_DIR/telemetry.json"

# Parse arguments
SINCE_FILTER=""
COMMAND_FILTER=""
OUTPUT_FILE=""

for arg in $ARGUMENTS; do
  case $arg in
    --since)
      shift
      SINCE_FILTER="$1"
      ;;
    --command)
      shift
      COMMAND_FILTER="$1"
      ;;
    --output)
      shift
      OUTPUT_FILE="$1"
      ;;
  esac
done

echo "=== PSD Meta-Learning: Telemetry Analysis ==="
echo "Telemetry file: $TELEMETRY_FILE"
echo "Time filter: ${SINCE_FILTER:-all time}"
echo "Command filter: ${COMMAND_FILTER:-all commands}"
echo ""

# Verify telemetry file exists
if [ ! -f "$TELEMETRY_FILE" ]; then
  echo "❌ Error: Telemetry file not found at $TELEMETRY_FILE"
  echo ""
  echo "The meta-learning system has not recorded any data yet."
  echo "Use workflow commands (/work, /test, etc.) to generate telemetry."
  exit 1
fi

Phase 2: Read and Validate Telemetry Data

Use the Read tool to examine the telemetry file structure:

# Read telemetry.json
cat "$TELEMETRY_FILE"

Expected structure:

{
  "version": "1.0.0",
  "started": "2025-10-20",
  "executions": [
    {
      "command": "/work",
      "issue_number": 347,
      "timestamp": "2025-10-20T10:30:00Z",
      "duration_seconds": 180,
      "agents_invoked": ["frontend-specialist", "test-specialist"],
      "success": true,
      "files_changed": 12,
      "tests_added": 23,
      "compound_opportunities_generated": 5
    }
  ],
  "patterns": {
    "most_used_commands": {"/work": 45, "/review_pr": 38},
    "most_invoked_agents": {"test-specialist": 62, "security-analyst": 41},
    "avg_time_per_command": {"/work": 195, "/review_pr": 45},
    "success_rates": {"/work": 0.94, "/architect": 0.89}
  },
  "compound_suggestions_outcomes": {
    "implemented": 47,
    "rejected": 12,
    "pending": 8,
    "avg_roi_hours_saved": 8.3
  }
}

Phase 3: Analyze Telemetry Data

Now analyze the data using extended thinking to detect patterns:

Analysis Tasks

  1. Activity Summary:

    • Count total executions (filtered by --since if specified)
    • Calculate most-used commands with percentages
    • Compute average time saved vs manual workflow
    • Track success/failure rates
  2. Pattern Detection:

    • Agent Correlation Analysis: Identify which agents frequently run together

      • Look for agent pairs appearing in >70% of executions together
      • Example: "security-analyst always precedes test-specialist (92% correlation)"
    • Time Bottleneck Analysis: Compare average durations

      • Identify operations taking 2-3x longer than average
      • Example: "PR reviews take 3x longer without code-cleanup first"
    • Bug Clustering: Analyze issue patterns

      • Look for similar error types occurring multiple times
      • Example: "UTF-8 bugs occurred 3 times in 2 months"
    • Workflow Inefficiencies: Find sequential operations that could be parallel

      • Detect commands always run in sequence
      • Calculate potential time savings
  3. Optimization Candidates:

    • Chain operations that always run together
    • Add validation steps that would prevent failures
    • Parallelize independent agent invocations
  4. Predictive Alerts:

    • Security Risk Patterns: Code changed frequently without security review

      • Example: "Auth code changed 7 times without security review → 82% probability of incident"
    • Performance Degradation: Metrics trending negatively

    • Technical Debt Accumulation: Patterns indicating growing complexity

Phase 4: Generate Analysis Report

Create a comprehensive markdown report with the following structure:

## TELEMETRY ANALYSIS - [Current Date]

### Activity Summary
- **Commands Executed**: [total] (this [period])
- **Most Used**: [command] ([percentage]%), [command] ([percentage]%), [command] ([percentage]%)
- **Avg Time Saved**: [hours] hours/[period] (vs manual workflow)
- **Overall Success Rate**: [percentage]%

### Patterns Detected

[For each significant pattern found:]

**Pattern #[N]**: [Description of pattern with correlation percentage]
   → **OPPORTUNITY**: [Specific actionable suggestion]
   → **IMPACT**: [Time savings or quality improvement estimate]

Examples:
1. **Security audits always precede test commands** (92% correlation)
   → OPPORTUNITY: Auto-invoke security-analyst before test-specialist
   → IMPACT: Saves 5min per PR by eliminating manual step

2. **PR reviews take 3x longer without code-cleanup first** (avg 45min vs 15min)
   → OPPORTUNITY: Add cleanup step to /review_pr workflow
   → IMPACT: Saves 30min per PR review (15 hours/month at current volume)

3. **UTF-8 bugs occurred 3 times in 2 months** (document processing)
   → OPPORTUNITY: Create document-validator agent
   → IMPACT: Prevents ~40 hours debugging time per incident

### Workflow Optimization Candidates

[List specific, actionable optimizations with time estimates:]

- **Chain /security_audit → /test**: Saves 5min per PR, eliminates context switch
- **Add /breaking_changes before deletions**: Prevents rollbacks (saved ~8hr last month)
- **Parallel agent invocation for independent tasks**: 20-30% time reduction in multi-agent workflows
- **Auto-invoke [agent] when [condition]**: Reduces manual orchestration overhead

### Predictive Alerts

[Based on patterns and thresholds, identify potential future issues:]

⚠️  **[Issue Type] risk within [timeframe]**
   → **CONFIDENCE**: [percentage]% (based on [N] similar past patterns)
   → **EVIDENCE**:
      - [Specific data point 1]
      - [Specific data point 2]
      - [Comparison to similar past issue]
   → **PREVENTIVE ACTIONS**:
      1. [Action 1]
      2. [Action 2]
   → **ESTIMATED COST IF NOT PREVENTED**: [hours] debugging time
   → **PREVENTION COST**: [hours] (ROI = [ratio]x)

### Trend Analysis

[If sufficient historical data exists:]

**Code Health Trends**:
- ✅ Technical debt: [trend]
- ✅ Test coverage: [trend]
- ⚠️  [Metric]: [trend with concern]
- ✅ Bug count: [trend]

### Recommendations

[Prioritized list of next steps:]

1. **IMMEDIATE** (High confidence, low effort):
   - [Suggestion]

2. **SHORT-TERM** (High impact, moderate effort):
   - [Suggestion]

3. **EXPERIMENTAL** (Medium confidence, needs A/B testing):
   - [Suggestion]

---

**Analysis completed**: [timestamp]
**Data points analyzed**: [count]
**Time period**: [range]
**Confidence level**: [High/Medium/Low] (based on sample size)

**Next Steps**:
- Review patterns and validate suggestions
- Use `/meta_learn` to generate detailed improvement proposals
- Use `/meta_implement` to apply high-confidence optimizations

Phase 5: Output Report

# Generate timestamp
TIMESTAMP=$(date "+%Y-%m-%d %H:%M:%S")

# If --output specified, save to file
if [ -n "$OUTPUT_FILE" ]; then
  echo "📝 Saving analysis to: $OUTPUT_FILE"
  # Report will be saved by the Write tool
else
  # Display report inline
  echo "[Report content displayed above]"
fi

echo ""
echo "✅ Analysis complete!"
echo ""
echo "Next steps:"
echo "  • Review patterns and validate suggestions"
echo "  • Use /meta_learn to generate detailed improvement proposals"
echo "  • Use /meta_implement to apply high-confidence optimizations"

Analysis Guidelines

Pattern Detection Heuristics

Strong Correlation (>85%):

  • Two events occur together in >85% of cases
  • Suggests causal relationship or workflow dependency
  • HIGH confidence for auto-implementation

Moderate Correlation (70-85%):

  • Events frequently associated but not always
  • Suggests common pattern worth investigating
  • MEDIUM confidence - good candidate for experimentation

Weak Correlation (50-70%):

  • Events sometimes related
  • May indicate contextual dependency
  • LOW confidence - needs human validation

Time Bottleneck Detection

Significant Bottleneck:

  • Operation takes >2x average time
  • Consistent pattern across multiple executions
  • Look for common factors (missing cleanup, sequential vs parallel, etc.)

Optimization Opportunity:

  • Compare similar operations with different durations
  • Identify what makes fast executions fast
  • Suggest applying fast-path patterns to slow-path cases

Predictive Alert Criteria

High Confidence (>80%):

  • Pattern matches ≥3 historical incidents exactly
  • Risk factors all present and trending worse
  • Generate specific preventive action plan

Medium Confidence (60-79%):

  • Pattern similar to 1-2 past incidents
  • Some risk factors present
  • Suggest investigation and monitoring

Low Confidence (<60%):

  • Weak signals or insufficient historical data
  • Mention as potential area to watch
  • Don't generate alerts (noise)

Empty or Insufficient Data Handling

If telemetry is empty or has <10 executions:

## TELEMETRY ANALYSIS - [Date]

### Insufficient Data

The meta-learning system has recorded [N] executions (minimum 10 required for meaningful analysis).

**Current Status**:
- Executions recorded: [N]
- Data collection started: [date]
- Time elapsed: [duration]

**Recommendation**:
Continue using workflow commands (/work, /test, /review_pr, etc.) for at least 1-2 weeks to build sufficient telemetry data.

**What Gets Recorded**:
- Command names and execution times
- Success/failure status
- Agents invoked during execution
- File changes and test metrics

**Privacy Note**: No code content, issue details, or personal data is recorded.

---

Come back in [X] days for meaningful pattern analysis!

Important Notes

  1. Statistical Rigor: Only report patterns with sufficient sample size (n≥5 for that pattern)
  2. Actionable Insights: Every pattern should have a concrete "OPPORTUNITY" with estimated impact
  3. Privacy: Never display sensitive data (code content, issue descriptions, personal info)
  4. Confidence Levels: Always indicate confidence based on sample size and correlation strength
  5. Time Periods: When using --since, clearly state the analysis window
  6. False Positives: Acknowledge when correlation might not equal causation
  7. ROI Focus: Estimate time savings/quality improvements in concrete terms (hours, bugs prevented)

Example Usage Scenarios

Scenario 1: Weekly Review

/meta_analyze --since 7d --output meta/weekly-analysis.md

Generates analysis of last week's activity, saved for review.

Scenario 2: Command-Specific Deep Dive

/meta_analyze --command work

Analyzes only /work command executions to optimize that workflow.

Scenario 3: Full Historical Analysis

/meta_analyze

Analyzes all telemetry data since system started.


Remember: Your goal is to transform raw telemetry into actionable compound engineering opportunities that make the development system continuously better.