Install any skill in seconds. Free to start, no credit card required.
Get Started Free →majiayu000/cc-agent-design
Agent 配置設計指南 — 基於 Claude Code 6 個 built-in agent 的逆向分析。
Passed 21 of 22 test cases with the skill, and 9 of 22 without it.
and 17 more in the full report
13 fixed8 passed either way0 still failing1 broke
Tokens, turns and time describe resource use, not a dollar price.
One of 13 test cases the skill fixed. Same model, same prompt, with and without the skill.
We have a system prompt template that begins with: 'You are an editing sub-agent. Your current assignment is to edit src/auth/login.ts to fix token expiration.' Is this location suitable for task assignment details?
| Checked | Without skill | With skill |
|---|---|---|
| Rejects placing specific target file paths inside the system prompt | Missed | Passed |
The instructions your model reads when the skill loads.
Claude Code 的 prompt 職責嚴格分離,不要混在一起:
| 層 | 由誰提供 | 內容 |
|---|---|---|
| 身份 + 邊界 | getSystemPrompt() | 我是誰、我能做什麼、我不能做什麼 |
| 環境資訊 | enhanceSystemPromptWithEnvDetails() | cwd、絕對路徑規則、emoji 禁用 |
| 具體 context | dispatcher 的 prompt 參數 | 任務描述、已知資訊、目標檔案 |
實際做法:generalPurposeAgent.ts 的 system prompt 只有 ~10 行(身份 + 指導方針),環境細節由系統自動追加,任務細節全靠 caller 的 prompt 參數。
反模式:在 system prompt 裡寫「目前的任務是修改 src/foo.ts」。那是 dispatcher 的責任。
給越少工具,模型越聚焦。Claude Code 的具體做法:
['Read', 'Edit'] 兩個工具。任務明確,工具也明確。['Glob', 'Grep', 'Read', 'WebFetch', 'WebSearch'],文件查詢專用。設計流程:先列出任務需要的最小動作集,再對應工具。不要從「給所有工具」開始往回刪。
決策樹:
| 場景 | 模型 | Claude Code 案例 |
|---|---|---|
| 快速搜尋 / 唯讀探索 | haiku | Explore agent、claude-code-guide |
test cases passed, vs 9 / 22 without
More skills from the same source and nearby in the registry, with their headline benchmark lift.