Local session logs used to diagnose AI coding tool degradation
What happened
A November 17 article in Computing Life · Share · Yage Research (鸭哥调研) describes how the .jsonl session logs kept by AI coding tools such as Claude Code can be used, without adding any instrumentation, to trace config changes, unusual tool use and drops in answer quality. It collects several user audit reports and lists metrics to analyze, including cached tokens, thinking-signature length, the read/write ratio, and the share of corrections in adjacent turns. One report records the user correction rate rising from 3.3% to 20.2%. What the article offers so far is a local-log-based troubleshooting method plus user audit cases.
Written by AI from the coverage · updated 3 days ago
Coverage
Follow the reports to see the story from different sides.
- Computing Life · Share · Yage如何用本地日志排查 Claude Code 等 AI 编程工具的能力退化
Claude Code 等 AI 编程工具留存的 .jsonl 会话日志,可用于排查配置变动、工具使用异常和回答质量退化,无需额外埋点。文章汇总多份用户审计报告,以缓存命中 token、思考签名长度、读写比和相邻轮次纠错比例作为分析指标,其中一份报告记录的用户纠错比例从 3.3% 升至 20.2%。
Heat over time
Heat now 3·Comparable peak 9(Nov 18 13:00)·Comparable change over 24 hours -49%
The trend only compares accounts observed without gaps, so its range may be smaller than the current heat. Hover or tap the chart for each hour; the left and right arrow keys step through it.