pattern-miner

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