Pattern Miner Skill

Overview

The pattern-miner Skill analyzes historical project data to identify recurring patterns, predict future issues, and provide data-driven recommendations for estimation and planning.

Purpose: Learn from past runs to make better predictions and avoid repeating mistakes.

When to Use

• After creating task groups - Adjust estimates based on historical patterns
• Before major features - Predict effort based on similar past features
• Weekly retrospectives - Identify systemic issues
• Quarterly planning - Understand team's actual velocity patterns

What It Analyzes

Pattern Types Detected

1. Task Duration Patterns

• "Database tasks always take 2.5x initial estimate" (85% confidence)
• "Authentication features require 3 revisions on average"
• "API integrations complete faster than expected (0.7x estimate)"

2. Module Risk Patterns

• "Payment module has 80% revision rate (high risk)"
• "Auth module always requires security review"
• "Frontend changes rarely need QA iterations"

3. Team Velocity Patterns

• "Team velocity increases 20% after first week of project"
• "Parallel mode with 3+ developers shows diminishing returns"
• "Friday deployments have 2x higher rollback rate"

4. Quality Patterns

• "Coverage drops correlate with rushed features (r=0.85)"
• "Security issues cluster in user input handling"
• "Lint violations spike during feature freeze"

Output Format

{
  "timestamp": "2024-11-08T12:00:00Z",
  "total_runs_analyzed": 25,
  "patterns_detected": [
    {
      "pattern_id": "db_task_overrun",
      "pattern": "database_tasks_overrun",
      "category": "estimation",
      "confidence": 0.85,
      "occurrences": 12,
      "description": "Database-related tasks take 2.5x longer than estimated",
      "evidence": {
        "avg_estimate": 5.0,
        "avg_actual": 12.5,
        "variance": 2.3
      },
      "recommendation": "Multiply database task estimates by 2.5x",
      "impact": "high"
    },
    {
      "pattern_id": "auth_security_review",
      "pattern": "auth_requires_security_review",
      "category": "process",
      "confidence": 1.0,
      "occurrences": 8,
      "description": "Authentication tasks always require security review (100%)",
      "recommendation": "Plan for security review in auth task timeline",
      "impact": "medium"
    }
  ],
  "lessons_learned": [
    "Payment processing features have 80% revision rate - break into smaller tasks",
    "Parallel mode with >3 developers shows coordination overhead (diminishing returns)",
    "Integration tests catch 90% of bugs that slip past unit tests"
  ],
  "predictions_for_current_project": [
    {
      "task_group": "C",
      "prediction": "Group C (payment processing) likely needs +30% time based on pattern 'payment_module_revisions'",
      "confidence": 0.78,
      "recommendation": "Add buffer to Group C estimate"
    }
  ],
  "estimation_adjustments": {
    "database_tasks": {
      "multiplier": 2.5,
      "confidence": 0.85
    },
    "authentication": {
      "multiplier": 1.5,
      "confidence": 0.90,
      "note": "Include security review time"
    },
    "api_integration": {
      "multiplier": 0.7,
      "confidence": 0.75,
      "note": "Usually faster than expected"
    }
  },
  "risk_indicators": [
    {
      "indicator": "revision_count > 3",
      "probability": 0.85,
      "outcome": "Requires tech lead escalation"
    },
    {
      "indicator": "story_points > 8",
      "probability": 0.70,
      "outcome": "Should be split into smaller tasks"
    }
  ]
}

Pattern Confidence Levels

Pattern Categories

1. Estimation - Task duration patterns
2. Process - Workflow patterns (e.g., always needs review)
3. Quality - Defect and revision patterns
4. Risk - Failure mode patterns
5. Team - Velocity and efficiency patterns

Usage Example

In PM agent:

# After creating task groups, refine estimates with patterns
/pattern-miner

# Read predictions
cat coordination/pattern_insights.json

# Apply adjustments
PATTERN_MULTIPLIER=$(jq -r '.estimation_adjustments.database_tasks.multiplier' coordination/pattern_insights.json)

if [ "$PATTERN_MULTIPLIER" != "null" ]; then
    # Adjust database task estimate
    NEW_ESTIMATE=$(echo "$ORIGINAL_ESTIMATE * $PATTERN_MULTIPLIER" | bc)
    echo "Adjusted database task estimate: $ORIGINAL_ESTIMATE → $NEW_ESTIMATE (${PATTERN_MULTIPLIER}x multiplier)"
fi

# Check predictions for current groups
jq -r '.predictions_for_current_project[] | "⚠️  \(.prediction)"' coordination/pattern_insights.json

Pattern Detection Algorithm

For each task type:
1. Group historical tasks by type (database, auth, API, etc.)
2. Calculate avg_actual_time / avg_estimated_time ratio
3. If ratio consistently >1.2 or <0.8 across ≥5 occurrences:
   → Pattern detected
4. Calculate confidence based on:
   - Number of occurrences (more = higher confidence)
   - Variance (lower = higher confidence)
   - Recency (recent data weighted more)

Benefits

✅ Data-driven estimates - Stop guessing, use historical data
✅ Avoid recurring mistakes - Learn from past patterns
✅ Proactive risk management - Predict issues before they happen
✅ Continuous improvement - Estimation accuracy improves over time
✅ Team-specific insights - Patterns unique to your team's velocity

Performance

• Execution time: 15-20 seconds
• Dependencies: jq (graceful fallback if not available)
• Minimum data: 5 historical runs for meaningful patterns
• Output: coordination/pattern_insights.json

Platform Support

• ✅ Linux/macOS (bash)
• ✅ Windows (PowerShell)

Related Skills

• velocity-tracker - Provides historical metrics input
• quality-dashboard - Uses pattern insights for predictions

Example Patterns

Real-world patterns detected:

Pattern: database_migrations_underestimated
Confidence: 0.88
Occurrences: 15/17 database tasks
Recommendation: Use 2.5x multiplier for DB tasks

Pattern: friday_deployments_risky
Confidence: 0.92
Occurrences: 11/12 Friday deploys had issues
Recommendation: Avoid Friday deployments

Pattern: parallel_mode_diminishing_returns
Confidence: 0.81
Occurrences: 8/10 runs with >3 developers
Recommendation: Cap parallelism at 3 developers