CI 日志分析
贴上失败的 CI 日志，几秒内看到根因，并得到可粘贴给编程助手的摘要。免费，无需注册，不会上传。
- 全程在浏览器中运行
- 无需注册
- 免费
Root-cause lines
Agent-ready summary — paste this back to your coding agent
No known failure pattern recognized
This log doesn't match any rule in the engine. Most frequent error-ish words below — no guessing, that's all the signal there is:
如何分析失败的 CI 日志
-
Paste the log
Drop in a failed run from GitHub Actions, GitLab, tsc, gcc, rustc, pytest, vitest or jest, go test, maven or gradle. Or open a .log file, or pick a real sample.
-
Autopsy
A deterministic rule engine strips timestamps, progress bars, and URL noise, then clusters the remaining error signals by type and weight. No LLM, no guessing.
-
Fix or hand off
Read the ranked root-cause lines with file and line numbers, or copy the agent-ready summary straight into your coding agent.
功能
-
Built for 4,000-line logs
Job and step tracking, noise classification, and signature clustering turn a wall of text into a ranked shortlist of the lines that actually explain the failure.
-
Deterministic, not AI
The same log always produces the same analysis. Rules are regex and table driven across the major CI ecosystems — no model calls, no drift.
-
Agent-ready summary
One copy gives you a markdown summary with verdict, stats, flaky signals, and the top root-cause lines, formatted to paste back into a coding agent.
-
Honest when it doesn't know
If nothing matches, it says so and shows the most frequent error-ish words — it never invents a diagnosis for a log it can't parse.
常见问题
Which CI systems and tools does it understand?
GitHub Actions and GitLab job structure and annotations, plus the error formats of tsc, gcc, rustc, pytest, vitest, jest, go test, maven, gradle, npm, pip, cargo, and more. Unknown formats fall through to an honest no-match card.
Is my log uploaded anywhere?
No. The analysis runs entirely in this browser tab, and the site's content security policy blocks outbound requests as a guarantee. Close the tab and the log is gone.
How big a log can it handle?
Up to 60 MB dropped as a file, though runs over about 20 MB get slow in the browser. For huge logs, trim to the failing region first — that is also faster to read.
What is the agent-ready summary?
A compact markdown block with the verdict, signal counts, flaky flags, and the top root-cause lines. Paste it to Claude Code, Codex, Cursor, or any coding agent as instant context.
Can it detect flaky tests?
Yes. It recognizes failed-then-passed runs, exhausted retries, and explicit flaky markers, and flags them separately from hard failures.
Is this free?
Yes — like every kitwisp tool: no sign-up, no limits on runs, no watermark on the output.