Red-builds study: GitHub Copilot cloud agent PRs

未关闭
#136 1 条评论 0 个 reaction 已指派 0 人 在 GitHub 查看

维护者通常 1 天内回复

还没有人认领这个 Issue。

评估

难度
4/5
预计耗时
3-5 天
新手友好度
45/100
Issue 类型
功能
描述清晰度
描述清楚
活跃度
活跃
技术栈
github, github-actions, typescript

调研方向

Read docs/red-builds.md and the Copilot integration work in #132, then inspect packages/diffci-cli/src/client/red-builds.ts and the diffci red-builds JSON output. Run the study using the read-only GitHub API and the specified CLI command, preserving denominators and separating Copilot PRs from all PRs. Done means a reviewed report with the requested per-repository table and notes on unrecognised job names.

由索引模型根据 Issue 内容生成。

描述

benchmark contribution help wanted integration

diffci red-builds (docs/red-builds.md) reads a repository's PR CI history and classifies each red build: could the change's own tests and checks have caught it before the push, or was it flaky, infrastructure, already broken on main, or CI-only? This study runs it on PRs opened by GitHub Copilot's cloud agent, to measure how many of an agent's red builds were preventable inside its own loop. It goes with the Copilot integration work in #132.

Method

  1. Find public repositories where Copilot's cloud agent opens PRs (head branches starting copilot/). Use the authenticated, read-only GitHub API only, one request at a time, at least 1.2 s apart, stopping below 1,000 remaining requests. Collect no personal data.
  2. Run diffci red-builds <owner/repo> --runs 200 --json (CLI 0.3.45 or later) on each repository, one at a time.
  3. Split the results into Copilot PRs (copilot/ branches, which AGENT_BRANCH in packages/diffci-cli/src/client/red-builds.ts already recognises) and all other PRs. If the JSON doesn't record each run's head branch, add that to the CLI first rather than inferring it.

Report

  • One row per repository: workflow, window, PR runs, red, preventable, partly, unconfirmed, shared (main already broken), CI-only, flaky, infra, and preventable runner-minutes. Show Copilot PRs and all PRs side by side.
  • Keep every denominator. Preventable counts are upper bounds: classification uses job types and failure annotations, not a DiffCI plan for each commit.
  • Note repositories where the job names weren't recognised and the run fell to CI-only.

Ground rules

  • Read-only. No comments, issues or PRs in the studied repositories as part of this study.
  • Results are published here only after a maintainer has reviewed the numbers.
  • No test replays: this reads CI history only.

Help wanted: if your own repository uses Copilot's cloud agent, you can run step 2 on it and post the table row below.

主要语言
TypeScript
星标
0
派生
4
平均合并
18 小时 41 分钟
30 天内合并 PR
48

环境准备

在 Codespaces 中打开

在浏览器里用你自己的 GitHub 账号启动这个项目的开发容器。

  • 没有 Dockerfile 或 Docker Compose 文件
  • 没有 Pull Request 模板
  • 阅读贡献指南

从这里开始

  1. 先读完整个 Issue,再读项目的贡献指南。
  2. 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
  3. Fork 仓库,在一个分支上完成修改。
  4. 提交 Pull Request,并在描述里引用这个 Issue 编号。

DiffCI/core 的其他 Issue

查看 DiffCI/core 的全部 Issue

相似的 Issue

更多 TypeScript Issue

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。