Back to Details

phargus-diagnose

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

Phargus Diagnose

Phargus 诊断

Explain the observed incident and choose a measurement that can distinguish its causes. Read data semantics and diagnosis. Use tools to capture or summarize the available source.
  1. Establish the affected process, symptom, interval and workload. Resolve the absolute directory containing this SKILL.md before invoking its scripts/phargus.mjs.
  2. Capture bounded SSE or read the supplied NDJSON. Run summarize and check invalid records, gaps, loss, resets and truncation before drawing conclusions. Live capture stops at 10,000 events by default (raise --max-events or shorten --duration-ms for a high-rate service); summarize reads files without an event-count cap.
  3. Compare affected and normal periods. Correlate request duration with event loop, GC, memory and backpressure. Retrieve raw data if SSE lacks necessary details.
  4. Separate measured observations from hypotheses. Inspect plausible application code only when evidence points there; preserve alternatives and uncertainty.
  5. Propose the smallest next experiment. When a fix is part of the user's request, check its prediction against a comparable before/after measurement.
Return values with units and timestamps, source/quality limits, ranked hypotheses and the next reproducible measurement. No data means insufficient evidence and a capture recipe. Read compatibility before assuming an Phargus capability.
Example request: "Requests became slow this morning; investigate this Phargus recording."
解释观察到的故障，并选择一种能够区分其成因的测量方式。 阅读数据语义和诊断说明。 使用工具捕获或汇总可用数据源。
  1. 确定受影响的进程、症状、时间区间和工作负载。调用 scripts/phargus.mjs 前，先解析包含此 SKILL.md 的绝对目录。
  2. 捕获有界的 SSE，或读取提供的 NDJSON。运行 summarize，并在得出结论前检查无效记录、数据缺口、丢失、重置和截断。实时捕获默认在 10,000 个事件时停止（对于高速率服务，可提高 --max-events 或缩短 --duration-ms）；summarize 读取文件时不受事件数量上限限制。
  3. 比较异常时段与正常时段。将请求耗时与 event loop、GC、内存和背压关联起来。如果 SSE 缺少必要细节，则获取原始数据。
  4. 区分实测观察与假设。仅当证据指向应用代码时才检查相关代码；保留其他可能性和不确定性。
  5. 提出最小化的下一步实验。如果用户请求中包含修复方案，请使用可比较的修复前后测量结果验证其预测。
返回带有单位和时间戳的数据、数据源与质量限制、按可能性排序的假设，以及下一项可复现的测量。没有数据意味着证据不足，应提供数据捕获方案。在假设 Phargus 具备某项能力前，请阅读兼容性说明。
示例请求：“今天早上请求变慢了；请调查这段 Phargus 录制数据。”