Hacktoberfest 2026:维护者为十月标记出来的 issue,仍然开放、适合新手。 浏览 Hacktoberfest issue

feat(evals): automated eval runner with hit, miss and false-positive scoring

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

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
56/100
Issue 类型
功能
描述清晰度
基本清楚
活跃度
活跃
领域
testing, tooling

调研方向

Start by reading the existing eval fixtures, their expected.md files, and evals/README.md; the issue does not name specific files beyond those paths. Define the machine-readable finding format and how fixture reports are scored, then add a runner that launches fixture sessions in parallel with a timeout. Done means it reports hit, partial, miss, and false positive totals, exits non-zero for missed required findings, and documents the runner while retaining the manual fallback.

由索引模型根据 Issue 内容生成。

描述

enhancement eval

What

Evals are run by hand today: one headless claude -p session per fixture, then each report is read against expected.md and scored by eye.

Why

Hand scoring is slow and drifts between runs, so a regression in SKILL.md or a reference can slip through a release.

Acceptance criteria

  • npm run eval (or node evals/run.mjs) starts one session per fixture in parallel with --plugin-dir and a timeout.
  • Each expected.md gains a machine-readable block (file, line range, grade, keywords) for must, should, acceptable and must-not findings.
  • The runner prints hit / partial / miss / false positive per fixture and a total, and exits non-zero when a required finding is missed.
  • evals/README.md documents the runner; the manual steps stay as a fallback.
主要语言
JavaScript
星标
0
派生
0
PR 合并指标
30 天内没有已合并 PR

环境准备

这个项目没有提供开发容器、Dockerfile 或贡献指南,环境需要你自己搭建:先看它的 README,通用步骤见我们的新手贡献指南。

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。

devmaxxx/stackwise 的其他 Issue

查看 devmaxxx/stackwise 的全部 Issue

相似的 Issue

更多 JavaScript Issue

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。