Add a skill-triggering eval set (prompt to expected skill and MCP tool)

未關閉
#55 0 則留言 0 個 reaction 已指派 0 人 在 GitHub 檢視

還沒有人認領這個 Issue。

評估

難度
3/5
預估耗時
1-2 天
新手友好度
58/100
Issue 類型
功能
描述清晰度
基本清楚
活躍度
活躍
技術堆疊
csharp, github-actions, json
領域
devtools, testing

研究方向

Start by reading validate_plugin.py and the existing CI checks for skill descriptions and MCP tool references. Add evals/skill-triggers.json with at least two prompts per skill, then extend validation to check that referenced skills and tools exist. Done means the eval set passes in CI; document how to run it manually and record the hit rate when skill descriptions change.

由索引模型根據 Issue 內容生成。

描述

good first issue skills testing

Why it matters

Skill descriptions were rewritten as "Use when ..." triggers (closes #42), and CI now checks the wording and that every MCP tool is mentioned by a skill. Nothing checks that a real user prompt loads the right skill or leads to the right tool call.

Suggested approach

  • Add evals/skill-triggers.json with 2 or 3 prompts per skill, for example {"prompt": "my game stutters every few seconds", "skill": "performance-optimization", "tool": "lookup_api"}.
  • Have validate_plugin.py check the file's structure (every skill and tool named exists, every skill has at least one prompt).
  • Document how to run the set manually in Cursor (or with claude plugin eval / another host if one supports Cursor-style skills), and record the hit rate in the PR when descriptions change.

Done when

Every skill has at least two eval prompts and the set is validated in CI.

主要語言
C#
星號
4
分支
0
PR 合併指標
30 天內沒有已合併 PR

環境準備

從這裡開始

  1. 先讀完整個 Issue,再讀專案的貢獻指南。
  2. 在 Issue 下留言說明你要接手 —— 這能避免兩個人做同樣的事。
  3. Fork 儲存庫,在一個分支上完成修改。
  4. 送出 Pull Request,並在描述裡引用這個 Issue 編號。

TMHSDigital/Unity-Developer-Tools 的其他 Issue

查看 TMHSDigital/Unity-Developer-Tools 的全部 Issue

相似的 Issue

更多 C# Issue

把新 issue 寄到你的電子郵件信箱

精選適合新手參與的 GitHub issue 摘要。