skill-lab / examples
CLAUDE.md0 starsChanged 10 days agoMITView on GitHub
風險分級，每次回覆第一行標出級別；只能升級，不能降級。bug 修復依改動範圍分級。gate 細節見 change-record skill。
In this repository20 other files it gives its agents.
- AGENTS.mdAGENTS.md
- api-and-interface-designSkill
- backlogSkill
- c4-architectureSkill
- change-recordSkill
- ci-cd-and-automationSkill
- competitive-briefSkill
- documentation-and-adrsSkill
- frontend-designSkill
- frontend-ui-engineeringSkill
- git-workflow-and-versioningSkill
- oss-release-readinessSkill
- performance-optimizationSkill
- security-and-hardeningSkill
- shipping-and-launchSkill
- synthesize-researchSkill
- user-researchSkill
- ux-copySkill
- webapp-testingSkill
- web-visual-designSkill
The file
Files
Skill 優先順序（使用者規則，優先於任何 skill）
風險分級，每次回覆第一行標出級別；只能升級，不能降級。bug 修復依改動範圍分級。gate 細節見 change-record skill。
| 級 | 判斷條件 | 流程 | 審查 |
| T0 瑣碎 | 只動文件或低風險設定，或純改名 | 直接改 | 自審 |
| T1 小修 | 最多改一個既有非測試原始碼檔、不新增原始碼檔、不改對外行為 | bug 修復先寫會失敗的測試 | 自審 |
| T2 功能 | 新行為、多檔改動，或碰到高風險設定或敏感路徑 | brainstorming bounded 路徑 → TDD | fresh-context reviewer，只給 diff 和需求，不附作者結論；每條結論附 path:line 或實際跑過的指令 |
| T3 架構 | 新專案或子系統、改公開介面、需要重組結構 | brainstorming 全流程 → spec → writing-plans → 子代理逐任務執行 | 每個 task 一次，最後整支 branch 的 final review |
- 被 gate 擋下時只能補齊或停下來問人；不得繞過、降級 Tier 或修改 gate 相關檔案。
- T3 的 spec 和 plan 寫在
docs/changes/。 - T2/T3 收尾時在最後列出「待人工確認」。
- 平行：互相獨立的唯讀任務可平行派發。會寫檔的實作代理一次只跑一個。
- 兩條規則衝突時，依本檔為準，並在回覆中說明採用了哪一條。
Skill 路由
- 需求釐清與規格只用 superpowers:brainstorming。市場與使用者研究用 competitive-brief、user-research、synthesize-research，結論寫回專案的 BACKLOG.md（backlog skill）。
- 架構圖用 c4-architecture，設計決策寫成 ADR（documentation-and-adrs）。
- 個人 skill 的原始檔在 skill-lab repo 的
skills/，改那裡再部署到~/.claude/skills，不要直接改部署後的副本。
環境
- Windows 11。Bash 工具是 Git Bash，另有 PowerShell；路徑用正斜線。
$HOME/mnt/...、/mnt/skills、/opt/pw-browsers是 Cowork Linux 沙盒路徑，在 Windows 上不存在，遇到就說明並跳過，不要硬跑。uv run一律加--isolated --no-project --python 3.13：否則會抓到上層目錄的.venv，其 Python 版本或套件可能不符。
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.