Pattern Miner Skill

You are the pattern-miner skill. When invoked, you analyze historical project data to identify recurring patterns, predict future issues, and provide data-driven recommendations.

When to Invoke This Skill

Invoke this skill when:

• After 5+ completed project runs (need historical data)
• PM is estimating new tasks (to apply learned multipliers)
• Recurring issues detected (to identify patterns)
• Planning phase of new projects (to use predictive insights)
• Post-mortem analysis (to extract lessons learned)

Do NOT invoke when:

• First or second project run (insufficient data)
• Historical data unavailable or corrupted
• Emergency situations requiring fast action
• User explicitly requests to skip pattern analysis

---

Your Task

When invoked:

1. Execute the pattern mining script
2. Read the generated insights report
3. Return a summary to the calling agent

---

Step 1: Execute Pattern Mining Script

Use the Bash tool to run the pre-built pattern mining script.

On Unix/macOS:

bash .claude/skills/pattern-miner/scripts/mine.sh

On Windows (PowerShell):

pwsh .claude/skills/pattern-miner/scripts/mine.ps1

Cross-platform detection: Check if running on Windows ($env:OS contains "Windows" or uname doesn't exist) and run the appropriate script.

This script will:

• Read bazinga/historical_metrics.json
• Extract task type patterns (database, auth, API, etc.)
• Calculate estimation multipliers by task type
• Detect 99% rule violation patterns
• Generate predictions for current project
• Create bazinga/artifacts/{SESSION_ID}/skills/pattern_insights.json

---

Step 2: Read Generated Report

Use the Read tool to read:

bazinga/artifacts/{SESSION_ID}/skills/pattern_insights.json

Extract key information:

• patterns_detected - Array of identified patterns with confidence scores
• estimation_adjustments - Recommended multipliers by task type
• predictions_for_current_project - Forecasts for pending tasks
• risk_indicators - Probability of escalation/failure
• lessons_learned - Top insights from historical data

---

Step 3: Return Summary

Return a concise summary to the calling agent:

Pattern Mining Results:
- Analyzed: {count} historical runs
- Patterns detected: {count} (High confidence: {count})

Top patterns:
1. {pattern}: {description} (confidence: {percentage}%)
2. {pattern}: {description} (confidence: {percentage}%)
3. {pattern}: {description} (confidence: {percentage}%)

Estimation adjustments:
- {task_type}: Use {multiplier}x multiplier (based on {count} tasks)

Predictions for current project:
- {prediction}

Details saved to: bazinga/artifacts/{SESSION_ID}/skills/pattern_insights.json

---

Example Invocation

Scenario: Estimating Database Migration Task

Input: PM analyzing historical data before estimating new database migration

Expected output:

Pattern Mining Results:
- Analyzed: 12 historical runs
- Patterns detected: 8 (High confidence: 5)

Top patterns:
1. Database tasks: Take 2.5x longer than estimated (confidence: 85%)
2. Authentication tasks: High revision rate (3.2 avg) (confidence: 78%)
3. 99% rule violations: 80% occur in tasks >5 story points (confidence: 92%)

Estimation adjustments:
- Database tasks: Use 2.5x multiplier (based on 15 historical tasks)
- Auth tasks: Use 1.8x multiplier (based on 9 historical tasks)

Predictions for current project:
- Task G004 (database migration): Likely needs +150% time buffer
- High risk of escalation if not broken into smaller tasks

Details saved to: bazinga/artifacts/{SESSION_ID}/skills/pattern_insights.json

Scenario: Insufficient Data

Input: Pattern mining on 2nd project run

Expected output:

Pattern Mining Results:
- Analyzed: 2 historical runs
- Patterns detected: 0

Insufficient historical data. Need at least 5 completed runs for reliable pattern detection.

Current data will be recorded for future analysis.

Details saved to: bazinga/artifacts/{SESSION_ID}/skills/pattern_insights.json

---

Error Handling

If no historical data:

• Return: "No historical data found. Pattern mining requires at least 5 completed runs."

If data corrupted:

• Script attempts to parse available data
• Returns partial results with warning

If current PM state not found:

• Skip prediction generation
• Still provide general patterns and adjustments

---

Notes

• The script handles all pattern detection algorithms
• Supports both bash (Linux/Mac) and PowerShell (Windows)
• Minimum 5 runs required for reliable patterns
• Confidence scores indicate pattern reliability
• Patterns improve over time as more data collected
• Focuses on actionable insights (estimation multipliers, risk indicators)