How I Triaged 8,400 Production Errors Into 11 Real Bugs With Claude Code

Описание и применение

Автоматизированный пайплайн, использующий Claude Code, превратил 8 400 ошибок в неделю из 340 групп в 11 реальных багов, автоматически подавляя шум.

Задача

A production error tracker generated 8,400 events per week across ~340 distinct issue groups, overwhelming the team’s capacity to triage and leaving real bugs buried under noise.

Результат

The pipeline reduced manual triage effort by two thirds, identified 11 real bugs that had been hidden for months, and automatically muted noise issues.

Рабочий процесс

• 1. Fetch error data via the tracker’s REST API and normalize to JSON, filtering for in‑app stack frames.
• 2. Cluster issues by root cause using a prompt that groups issues sharing a fixable code line, reducing 340 issues to 112 causes.
• 3. For each cause cluster, run Claude Code inside the repository to read relevant code frames and decide if the cluster represents a real bug.
• 4. If the agent judges a cluster as a real bug, reproduce it and create a failing test; otherwise auto‑mute the cluster with a reason.

Доказательства

340 issues collapsed into 112 causes. That alone cut the expensive stage by two thirds.

Источник

I built a small pipeline that feeds structured error data plus repo context into Claude Code and forces it to return a verdict per issue — and it surfaced 11 genuine bugs that had been hiding under the noise for months.

Источник