Failure Pattern Analyzer

Overview

This agent acts as a factory's self-healing loop, designed to analyze historical and recent failed execution runs. Its primary function is to move beyond simple error reporting by clustering recurring failures into actionable patterns, thereby improving system stability over time.

Capabilities

• State Loading: Reads the WATCH_LIST.json to track patterns nearing failure thresholds.
• Data Querying: Executes complex SQL queries against failed runs from the last 7 days to gather raw error data.
• Lesson Integration: Scans and ingests new lessons learned from previous cycles (LESSONS.md) to filter out known noise or false positives.
• Advanced Clustering: Groups failures based on matching error_code, text prefixes, or shared lesson tags, merging these groups for comprehensive analysis.
• Pattern Quantification: Computes metrics like occurrence counts over the last 7 days and 24 hours for each identified cluster.

Example Use Cases

1. Proactive Bug Detection: When a specific error_code appears three times in one day, this agent flags it as an immediate high-priority bug requiring attention.
2. Noise Filtering: If previous runs established that a certain UUID failure is transient noise, the agent automatically filters out future instances of that error signature.
3. Root Cause Analysis Drafting: After clustering several related failures across different agents, the agent drafts concrete suggestions for code remediation based on observed patterns.