What it does
Group failed GitHub Actions runs by root cause; separate real regressions from flaky failures.
- Connects 在你的电脑上运行
- Account or key No key needed
- Category 开发者工具
可搭配使用
教 AI 做这类工作的 Skill,可与本项一起使用
andrej-karpathy-skills
A single CLAUDE.md file to improve Claude Code behavior, derived from Andrej Karpathy's observations on LLM coding pitfalls.
graphify
Turn any codebase, with its docs, SQL schemas, configs, and PDFs, into a queryable knowledge graph. A /graphify skill for Claude Code, Cursor, Codex, and Gemini CLI: local deterministic AST parsing, every edge explained, no vector store.
为什么排在这里
列表按使用分(满分 100)排序,由以下部分组成
| 热度 | 0/80 | GitHub 星标或每周下载量:较大的计 60,较小的计 20 |
|---|---|---|
| 新鲜度 | 12/12 | 30 天内有发布或代码更新得满分 |
| 仍在响应 | 0/8 | 上次检查时云端端点有响应 |
| 使用分 | 12/100 |
质量: 基础检查通过 3/3 项(仅展示,不用于排序) · 上升势头: 暂无数据 (只用于“上升中”标签)
安装
选择你使用的 AI 应用,然后复制设置。
用一句话让 AI 帮你安装
请帮我安装 ci-triage。按照 https://mcp.popnx.com/i/wjddidi-1-ci-triage.md 操作,先把命令给我看,等我同意后再运行。粘贴到能在你电脑上工作的 AI(如 Claude Code、Cursor)。它会读取我们的步骤,先给你看命令,等你同意。无需密钥,无需注册。 · 查看 AI 会读取的全文
通过 3/3 项检查 · 在你的电脑上运行 · ✓ 命令已检查 2026年11月16日
第一步:你用哪个 AI?
选择你用的应用,步骤会显示在下方。
可连接 Claude,但需要先准备
需要先在电脑上安装 Node.js(只需一次) · 仅限电脑上的 Claude 应用(Windows 或 Mac),网页和手机不可用
让 AI 帮你安装
把这段话复制到 Claude 的对话里。AI 会先检查电脑,告诉你要改哪个文件,并且每次都等你同意。
请在这台电脑上为 Claude 安装 MCP server ci-triage,按顺序:1) 用 node -v 检查是否有 Node.js。没有的话,请让我从 https://nodejs.org/en/download 安装,不要自己安装。2) 只使用这份配置(已通过 PopNx MCP 检查):{ "mcpServers": { "ci-triage": { "command": "npx", "args": [ "-y", "ci-triage-cli" ] } } } 3) 告诉我要修改的文件完整路径或要运行的命令,等我同意。4) 不需要密钥。5) 按约定执行,然后确认 ci-triage 的工具真的出现了。这段话不包含你的密钥。如果需要密钥,AI 会请你自己粘贴到文件或应用的输入框里。
安装 Node.js(如果还没有)
已经装好了吗?先检查 · 打开终端,输入这条命令后按 Enter
node -v- 显示版本号(例如 v26.13.1)= 已安装,可跳过这一步
- 显示 "command not found"(Windows 显示 "is not recognized")= 还没安装,或刚装好但需要关掉终端再重新打开
怎么打开终端
Mac:按 ⌘ + Space,输入 Terminal 后按 Enter
Windows:打开“开始”菜单,输入 Windows PowerShell,然后选“打开”
还没安装: 打开官方下载页,选择 LTS 版本和点选式安装程序:Mac 用 .pkg · Windows 用 .msi。双击后一路下一步即可,不需要 curl 命令
装好后关掉终端再打开,重新运行同一条命令,应显示版本号:
node -v安装一次,所有需要 Node.js 的工具都能用
打开电脑版 Claude 应用的设置文件
SettingsDeveloperEdit Config
粘贴这些设置并保存文件
{ "mcpServers": { "ci-triage": { "command": "npx", "args": [ "-y", "ci-triage-cli" ] } } }退出 Claude 再重新打开
试一试
用 ci-triage 告诉我你能帮我做什么,然后做一件给我看✓ 如果 Claude 说正在使用 ci-triage,或回答中出现 ci-triage 的工具名称,就表示连接成功
菜单名称已于 2026年11月11日 对照 Anthropic 帮助页面核实 · 你的界面可能因版本略有不同
ChatGPT 暂时无法连接 ci-triage
ci-triage 需要安装在你的电脑上,而 ChatGPT 只能连接在网上运行的服务器
你可以改用这些方式
A · 使用能连接 ci-triage 的应用
- 需要先在电脑上安装 Node.js(只需一次) · 仅限电脑上的 Claude 应用(Windows 或 Mac),网页和手机不可用
- 需要先在电脑上安装 Node.js(只需一次)
- 需要先在电脑上安装 Node.js(只需一次)
- 需要先在电脑上安装 Node.js(只需一次)
B · 开发者工具 类中能连接 ChatGPT 的其他服务器
从同类中使用较多的里挑选,仅限用地址即可连接、无需密钥的
- Sentry✓ ChatGPT
- Compiler Explorer✓ ChatGPT
- GitDiagram✓ ChatGPT
可连接 Claude Code,但需要先准备
需要先在电脑上安装 Node.js(只需一次)
让 AI 帮你安装
把这段话复制到 Claude Code 的对话里。AI 会先检查电脑,告诉你要改哪个文件,并且每次都等你同意。
请在这台电脑上为 Claude Code 安装 MCP server ci-triage,按顺序:1) 用 node -v 检查是否有 Node.js。没有的话,请让我从 https://nodejs.org/en/download 安装,不要自己安装。2) 只使用这份配置(已通过 PopNx MCP 检查):claude mcp add --scope user ci-triage -- npx -y ci-triage-cli 3) 告诉我要修改的文件完整路径或要运行的命令,等我同意。4) 不需要密钥。5) 按约定执行,然后确认 ci-triage 的工具真的出现了。这段话不包含你的密钥。如果需要密钥,AI 会请你自己粘贴到文件或应用的输入框里。
安装 Node.js(如果还没有)
已经装好了吗?先检查 · 打开终端,输入这条命令后按 Enter
node -v- 显示版本号(例如 v26.13.1)= 已安装,可跳过这一步
- 显示 "command not found"(Windows 显示 "is not recognized")= 还没安装,或刚装好但需要关掉终端再重新打开
怎么打开终端
Mac:按 ⌘ + Space,输入 Terminal 后按 Enter
Windows:打开“开始”菜单,输入 Windows PowerShell,然后选“打开”
还没安装: 打开官方下载页,选择 LTS 版本和点选式安装程序:Mac 用 .pkg · Windows 用 .msi。双击后一路下一步即可,不需要 curl 命令
装好后关掉终端再打开,重新运行同一条命令,应显示版本号:
node -v安装一次,所有需要 Node.js 的工具都能用
打开终端
Mac:按 Command+空格,输入 Terminal · Windows:打开 PowerShell
粘贴这条命令并按 Enter
claude mcp add --scope user ci-triage -- npx -y ci-triage-cli在 Claude Code 中输入 /mcp
应能看到 ci-triage 在列表中并显示已连接
试一试
用 ci-triage 告诉我你能帮我做什么,然后做一件给我看✓ 如果 Claude Code 说正在使用 ci-triage,或回答中出现 ci-triage 的工具名称,就表示连接成功
菜单名称已于 2026年11月11日 对照 Anthropic 帮助页面核实 · 你的界面可能因版本略有不同
可连接 Cursor,但需要先准备
需要先在电脑上安装 Node.js(只需一次)
让 AI 帮你安装
把这段话复制到 Cursor 的对话里。AI 会先检查电脑,告诉你要改哪个文件,并且每次都等你同意。
请在这台电脑上为 Cursor 安装 MCP server ci-triage,按顺序:1) 用 node -v 检查是否有 Node.js。没有的话,请让我从 https://nodejs.org/en/download 安装,不要自己安装。2) 只使用这份配置(已通过 PopNx MCP 检查):{ "mcpServers": { "ci-triage": { "command": "npx", "args": [ "-y", "ci-triage-cli" ] } } } 3) 告诉我要修改的文件完整路径或要运行的命令,等我同意。4) 不需要密钥。5) 按约定执行,然后确认 ci-triage 的工具真的出现了。这段话不包含你的密钥。如果需要密钥,AI 会请你自己粘贴到文件或应用的输入框里。
安装 Node.js(如果还没有)
已经装好了吗?先检查 · 打开终端,输入这条命令后按 Enter
node -v- 显示版本号(例如 v26.13.1)= 已安装,可跳过这一步
- 显示 "command not found"(Windows 显示 "is not recognized")= 还没安装,或刚装好但需要关掉终端再重新打开
怎么打开终端
Mac:按 ⌘ + Space,输入 Terminal 后按 Enter
Windows:打开“开始”菜单,输入 Windows PowerShell,然后选“打开”
还没安装: 打开官方下载页,选择 LTS 版本和点选式安装程序:Mac 用 .pkg · Windows 用 .msi。双击后一路下一步即可,不需要 curl 命令
装好后关掉终端再打开,重新运行同一条命令,应显示版本号:
node -v安装一次,所有需要 Node.js 的工具都能用
在 Cursor 中以 Agent 模式打开 AI 聊天
试一试
用 ci-triage 告诉我你能帮我做什么,然后做一件给我看✓ 如果 Cursor 说正在使用 ci-triage,或回答中出现 ci-triage 的工具名称,就表示连接成功
菜单名称已于 2026年11月11日 对照 Cursor 帮助页面核实 · 你的界面可能因版本略有不同
可连接 VS Code,但需要先准备
需要先在电脑上安装 Node.js(只需一次)
让 AI 帮你安装
把这段话复制到 VS Code 的对话里。AI 会先检查电脑,告诉你要改哪个文件,并且每次都等你同意。
请在这台电脑上为 VS Code 安装 MCP server ci-triage,按顺序:1) 用 node -v 检查是否有 Node.js。没有的话,请让我从 https://nodejs.org/en/download 安装,不要自己安装。2) 只使用这份配置(已通过 PopNx MCP 检查):{ "servers": { "ci-triage": { "type": "stdio", "command": "npx", "args": [ "-y", "ci-triage-cli" ] } } } 3) 告诉我要修改的文件完整路径或要运行的命令,等我同意。4) 不需要密钥。5) 按约定执行,然后确认 ci-triage 的工具真的出现了。这段话不包含你的密钥。如果需要密钥,AI 会请你自己粘贴到文件或应用的输入框里。
安装 Node.js(如果还没有)
已经装好了吗?先检查 · 打开终端,输入这条命令后按 Enter
node -v- 显示版本号(例如 v26.13.1)= 已安装,可跳过这一步
- 显示 "command not found"(Windows 显示 "is not recognized")= 还没安装,或刚装好但需要关掉终端再重新打开
怎么打开终端
Mac:按 ⌘ + Space,输入 Terminal 后按 Enter
Windows:打开“开始”菜单,输入 Windows PowerShell,然后选“打开”
还没安装: 打开官方下载页,选择 LTS 版本和点选式安装程序:Mac 用 .pkg · Windows 用 .msi。双击后一路下一步即可,不需要 curl 命令
装好后关掉终端再打开,重新运行同一条命令,应显示版本号:
node -v安装一次,所有需要 Node.js 的工具都能用
在 VS Code 中以 Agent 模式打开 AI 聊天
试一试
用 ci-triage 告诉我你能帮我做什么,然后做一件给我看✓ 如果 VS Code 说正在使用 ci-triage,或回答中出现 ci-triage 的工具名称,就表示连接成功
菜单名称已于 2026年11月11日 对照 VS Code 帮助页面核实 · 你的界面可能因版本略有不同
开发者专用:终端命令和 JSON 设置
安装在电脑上
Claude Code
claude mcp add --scope user ci-triage -- npx -y ci-triage-cliClaude Desktop (claude_desktop_config.json)
{
"mcpServers": {
"ci-triage": {
"command": "npx",
"args": [
"-y",
"ci-triage-cli"
]
}
}
}Cursor (~/.cursor/mcp.json)
{
"mcpServers": {
"ci-triage": {
"command": "npx",
"args": [
"-y",
"ci-triage-cli"
]
}
}
}VS Code (mcp.json)
{
"servers": {
"ci-triage": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"ci-triage-cli"
]
}
}
}这些设置是根据作者在注册中心的条目生成的。如果 README 的写法不同,请以 README 为准。
一览
哪些应用可以使用它
依据这个服务器的连接方式和各应用自己的文档,检查于 2026年11月10日。
在你的电脑上运行,适用于
ClaudeGooseLM StudioMsty StudioCherry StudioCursorVS Code + GitHub CopilotVisual StudioWindsurf (Devin Desktop)ZedKiroClineContinueJetBrains AI AssistantClaude CodeCodex CLIGemini CLIOpenCodenodecodeWarpLibreChatRaycast
它能做什么
它在你的电脑上运行;安装后,工具列表会显示在你的 AI 应用中。
基础检查i
我们检查的是任何人都能看到的信息;我们没有阅读代码。不适用的检查项不计入总数。 · 检查于 2026年11月19日
- ✓ 通过
作者与仓库一致
注册名称已通过 GitHub 账户 WJDdidi-1 验证,该账户拥有此仓库
- ✓ 通过
仍在维护
代码最后更新于今天
- ✓ 通过
有许可证
使用 Apache-2.0
- – 不适用
端点有响应
它在你的电脑上运行,没有互联网端点
- – 不适用
使用 OAuth 登录
在你的电脑上运行,改用本地设置
- … 尚未检查
工具描述中没有隐藏指令
它在你的电脑上运行;我们只能读取互联网端点的工具列表
已公开的漏洞(OSV.dev)
尚未检查
作者:申请重新检查或报告信息错误
给作者:在 README 中加徽章
如果你是作者,可把这段代码贴到 README。徽章会链接回本页。
[](https://mcp.popnx.com/en/servers/wjddidi-1-ci-triage)讨论
提问、分享技巧或报告无法使用。按“有帮助”数量排序。
用户评价
还没有评价
与“开发者工具”中的其他服务器比较i
- 星标数多于此分类中其他 6182 个服务器的 0%
- 发布到注册中心的时间比此分类中其他 6182 个服务器的 91% 更新
- 基础检查结果等于或优于此分类中其他 6182 个服务器的 99%
与注册中心里这个分类中所有活跃的服务器比较。只有当发布者拥有该仓库时,星标数才计入。
常见问题
ci-triage 安全吗?
它在 3 项适用的基础检查中通过了 3 项。我们没有阅读代码。在添加密钥或令牌之前,请先看看这个服务器要求哪些权限,并使用只具备所需权限的密钥。
它需要哪些密钥或代码?
注册中心没有列出必需的密钥,不过有些服务器会在首次使用时要求你登录。
它在我的电脑上运行还是在云端?
它在你的电脑上运行。你的 AI 应用会在需要时启动它,数据不会经过作者的服务器(除非它自己调用在线服务)。
它还在维护吗?
GitHub 仓库最后更新于今天,最新版本(0.1.2)于3天前进入注册中心。
哪些应用可以使用它?
支持 MCP 的应用,例如 Claude Desktop、Claude Code、Cursor 和 VS Code。你的电脑上还需要有 Node.js。
有什么可以替代它?
同一分类中有:GitDiagram, Appllama, Cloudflare MCP, Omnara, Repowise
功能相近的其他选择
同一分类、其他作者、命令已通过检查。官方的排在前面,再按星标排序。
如何看卡片:↓ 每周下载量i★ GitHub 星标i三大支柱:评价 · 检查 · 使用情况i更新 = 最新版本发布到注册中心的时间i
GitDiagram
Architecture diagrams, explanations and Mermaid source for public GitHub repositories.
Appllama
Study screens, flows and paywalls from top-earning iOS apps, then build from what wins.
Omnara
Deploy durable AI agents with Omnara, an open-source managed-agent platform.
Repowise
Codebase intelligence for AI coding agents: graph, git history, docs, decisions, code health.
| 本项 ci-triage | GitDiagram | Appllama | Cloudflare MCP | Omnara | Repowise | |
|---|---|---|---|---|---|---|
| 运行位置 | 需要安装在电脑上 | 在云端运行,无需安装 | 在云端运行,无需安装 | 在云端运行,无需安装 | 在云端运行,无需安装 | 云端或安装在电脑上都可以 |
| 需要密钥 | 否 | 否 | 否 | 否 | 否 | 否 |
| 可在 Claude 中按 + 添加 | 否 | 否 | 否 | 是 | 否 | 否 |
| 基础检查 | 3/3 | 4/4 | 4/4 | 4/4 | 4/4 | 5/5 |
| 许可证 | Apache-2.0 | MIT | MIT | Apache-2.0 | Apache-2.0 | AGPL-3.0 |
| 价格 | 登记中未列出费用 | 登记中未列出费用 | 登记中未列出费用 | 登记中未列出费用 | 登记中未列出费用 | 登记中未列出费用 |
作者的 README
打开 README
按原文从作者的仓库复制。PopNx MCP 没有审查或背书其内容。HTML、图片和脚本已被移除。链接会离开本站。
ci-triage
CI is red. Which failure matters, and is it a real regression or just flaky?
ci-triage answers that in one page: it fetches a repo's recent failed GitHub Actions runs, reads the failed job logs, groups failures by root cause, and tells you which ones are real regressions and which are flaky — so you fix the one thing that matters instead of reading thousands of log lines by hand.
Before / after
Before: CI is red on 3 runs. You open each run, find the failed job, scroll through the log, and try to remember whether you saw this error yesterday.
After:
$ ci-triage acme/widgets
ci-triage — acme/widgets Window: 2 recent runs listed, 2 failed examined Root causes: 1 (1 real, 0 flaky)
#1 test · 2 runs · real regression FAIL src/components/Button.test.tsx 3 test-runner failure signals in the error lines. fails on every recent run — looks like a real regression, not flakiness. first seen 2026-11-12 10:00 · last seen 2026-11-12 14:03 runs: #2 (test), #1 (test) evidence: FAIL src/components/Button.test.tsx ● Button › renders the disabled state expect(received).toBe(expected) // Object.is equality
Overall: 1 root cause — 1 real regression, 0 flaky
One root cause, two runs, real regression, with the failing test named. That is the whole triage.
Install
npm install -g ci-triage-cli
Requires Node 18 or newer. No other dependencies — pure Node.js standard library.
Rather not install it by hand? Paste this to your coding agent: "Install ci-triage (npm install -g ci-triage-cli if it's not on PATH), then run ci-triage --skill-generate so you triage red CI before reading any log."
Use
ci-triage <owner/repo> # triage the last 10 failed runs
ci-triage acme/widgets --runs 3 # fewer runs, faster ci-triage acme/widgets --workflow ci.yml --branch main ci-triage acme/widgets --json # machine-readable output ci-triage acme/widgets --top 3 # show only the top 3 root causes ci-triage acme/widgets --report triage.md # Markdown report to paste into a PR, an issue, or an agent chat ci-triage acme/widgets --annotations # also emit ::error:: / ::warning:: annotations for GitHub Actions
--report writes the same Markdown the GitHub Action puts in the Step Summary — the ranked report, not machine JSON — so the file, the summary, and the terminal output never disagree. Under --json the confirmation goes to stderr, so stdout stays parseable. --annotations does the same job for the run's annotations panel: one workflow-command line per root cause (::error:: for a real regression, ::warning:: for flaky or unknown), capped at 10 per level — GitHub's own limit. When the log names the failing file with a line (src/utils/format.ts:18:7, a Jest stack frame), the annotation is pinned to that file and line, so it also shows in the PR's changed-files view, one click from the code; a bare filename, an absolute runner path, or a node_modules frame stays unpinned rather than pointing at the wrong file. Under --json the lines go to stderr, so stdout stays parseable.
Exit codes: 0 when no failed job turned up, or when every failure in the window is a recorded known failure; 1 when failures were found (including failures whose logs could not be read — the repo is not clean), 2 for a usage error or an API error.
Known failures
A failure the team already knows about and is living with should not re-alert as a real regression on every triage — it drowns the new regressions. Record it once, then triage against the record:
ci-triage acme/widgets --write-known .ci-triage-known-failures.json
ci-triage acme/widgets --known .ci-triage-known-failures.json
--write-known records the current clusters and merges with entries already in the file; --known marks the clusters the file covers as known failures instead of new regressions. Matching is by failure signature, so a known failure stays known when it moves to another workflow. The file holds signature hashes and headlines only, so it is safe to commit. A recorded failure that stops appearing is listed as "not seen in this window (possibly fixed)" — remove it from the file if it is fixed. The method is Playwright's test.fail(): a known bug that is still exercised, expected to fail, and noticed when it stops failing. In the Action, a triage with only known failures passes the step instead of failing it.
Shell completion for bash, zsh and fish: ci-triage completion bash prints a completion script generated from the same option list the CLI parses, so it can never suggest a flag the CLI rejects. Install it with, e.g. ci-triage completion bash > ~/.bash_completion.d/ci-triage, or source <(ci-triage completion bash) for a one-off session. Mistyped a flag anyway? The CLI suggests the closest match (Did you mean ...?) instead of a bare error.
Auth — read this first
Set GH_TOKEN (or GITHUB_TOKEN). Without a token the triage can list the failed runs and jobs, but it cannot download their logs: GitHub does not allow anonymous job-log downloads, even for public repos. The first refused download stops the rest — the refusal applies to every job, so the triage lists all the failed jobs with the same reason instead of spending the anonymous budget on downloads known to fail. Without a token you get an honest report instead of a triage:
No clusters: 9 failed jobs found, but their logs could not be read.
Failed jobs #41–#49 · 2026-11-14 12:03 – 2026-11-15 08:41 Reason: downloading this job log needs a GitHub token (403) — GitHub does not allow anonymous log downloads; set GH_TOKEN or run gh auth login Overall: UNKNOWN — 9 failed jobs found, logs unreadable
With a token you get 5,000 API requests per hour. gh auth login works too when the GitHub CLI is logged in.
ci-triage doctor checks the setup itself before you triage anything: Node version, ci-triage on PATH, the home directory --skill-generate writes into, installed skills (and whether their frontmatter is valid — a broken one silently never triggers), whether a token is available, and whether the GitHub API answers (one cheap call, which also reports the rate-limit budget left this hour). It is read-only — it changes nothing — and every failure comes with a copy-paste fix. Warnings are for optional wiring (a missing skill, no token, an unreachable API); only failures make it exit 1. --json gives the same checklist as data. The method is Claude Code's claude doctor: one command that answers "is my setup healthy?" instead of debugging each piece by hand.
How it decides
- Signature: the error lines of a failed job, with timestamps, ANSI codes, file paths, hashes, versions and counts stripped — so the same failure on two runs has the same signature even when the paths and timings differ.
- Clustering: failures with the same signature are one root cause, ranked by how many runs they hit.
- Classification: each cluster is a test failure, a build failure, or an infra failure (timeouts and the like), from the error lines themselves.
- Flaky vs real: a cluster is flaky when the same commit passed on a later run, or when failures and successes alternate in the window. Failing on every recent run is a real regression.
- Known failures: a cluster whose signature is recorded in a known-failures file (
--known) is reported as a known failure, not a new regression. A triage with only known failures exits 0. - Regression window: a real regression names the last green run and the first red one ("broke between #1 (ccc0000) and #2 (aaa0000)");
git log ..lists the suspect commits. The method is git bisect's — the good and bad endpoints — answered from the run history instead of a binary search, since CI already ran every commit's build. The report also lists the files that changed between those two commits (GitHub's compare endpoint, capped at 10 with a count of the rest), so the fix starts from the suspect files, not just the suspect range; a compare that fails leaves the window line as it was. A flaky failure has no window; when no passing run precedes the first failure, the report says the regression predates the runs listed.
Logs expire: GitHub keeps job logs for 90 days. A failed job whose log is gone is recorded, not fatal.
For agents
ci-triage --mcp # stdio MCP server: triage_repo, triage_log
ci-triage --skill-generate # write a SKILL.md for this project (never overwrites) and a short section in AGENTS.md ci-triage --skill-generate --agents # same skill, in .agents/skills/ — the shared location Gemini CLI and Codex discover skills from
triage_log classifies a pasted log without touching the network — useful when you already have the log open. Both tools return the same JSON the CLI's --json produces, as the text block and again as structuredContent, the MCP structured-output channel (each tool declares a loose outputSchema: { type: 'object' }; error results carry none), so a calling application can read the fields typed instead of parsing the text.
The MCP server also serves one prompt, triage_ci (prompts/list / prompts/get): the one-shot "what broke in CI and what do I fix first?" workflow — call triage_repo with the repo (and workflow, if given), report the root causes ranked real regressions first, then flaky, then unknown, then known failures, and do not re-run or fix anything without the user agreeing — for MCP clients that let the user pick a prompt instead of typing a request. The method is FastMCP's: prompts are the third MCP primitive, the user-invoked template next to tools and resources. triage_repo also takes knownFile: a known-failures file to mark recorded clusters as known instead of new regressions.
Claude Code users can install the skill without installing the CLI first: /plugin marketplace add WJDdidi-1/ci-triage, then /plugin install ci-triage@ci-triage. The skill lives in this repo at skills/ci-triage/SKILL.md, so skill indexes can find it too.
Claude Desktop: the repo ships a manifest.json (MCPB, the renamed DXT format) — pack it with npx @anthropic-ai/mcpb pack and drag the .mcpb file onto Settings → Extensions. The bundle is just the server (zero dependencies, so there are no node_modules to bundle), and .mcpbignore keeps tests and tooling out of it.
GitHub Action
Use it from a workflow (it needs a token with actions:read on the repo; the workflow's own GITHUB_TOKEN can read its own repo's runs):
jobs:
triage: runs-on: ubuntu-latest steps:
- uses: WJDdidi-1/ci-triage@main with: repo: ${{ github.repository }} runs: 10
The action fails the job when it finds root causes, so the triage report lands in the log and in the run's Step Summary. Add --annotations to surface each root cause on the run's annotations panel too — one ::error:: per real regression, ::warning:: for flaky, unknown, or known — so the cause is visible without opening the log at all. It needs no token and no file, and it works on fork PRs, where the Checks API cannot write; the same trick reviewdog's github-annotations reporter uses. Locally it is a no-op: the lines only become annotations when a step runs in GitHub Actions.
How fast is it
On a real repo with 3 failed runs and 9 failed jobs, the triage took about 8 seconds wall-clock — against opening 9 job logs by hand and reading them. The clustering and classification themselves run in milliseconds; the time is the GitHub API round-trips.
Development
npm test # node:test, no extra dependencies, run serially
node scripts/bump-version.mjs # bump every version pin at once
Coverage uses Node's built-in runner (Node 22+ only — the suite itself runs on Node 18+): `node --test --test-concurrency=1 --experimental-test-coverage --test-coverage-include='lib/**/*.mjs' --test-coverag