feat(contributors): add fishtvlvoe - 10 lessons on multi-agent delegation and SDD workflow - #1
Conversation
…tion and SDD workflow
There was a problem hiding this comment.
Pull request overview
Adds a new contributor profile (fishtvlvoe) and publishes a set of workflow/verification lessons to the contributors knowledge base, plus registers the contributor in the machine-readable index.
Changes:
- Added
contributors/fishtvlvoe/index.mdcontributor profile page - Added 10 lesson pages under
contributors/fishtvlvoe/(001–010) - Updated
_data/contributors.ymlto includefishtvlvoe
Reviewed changes
Copilot reviewed 12 out of 12 changed files in this pull request and generated 4 comments.
Show a summary per file
| File | Description |
|---|---|
| contributors/fishtvlvoe/index.md | New contributor landing page listing the 10 lessons |
| contributors/fishtvlvoe/001-fallback-before-giving-up.md | Lesson 001 content (tool fallback discipline) |
| contributors/fishtvlvoe/002-report-without-being-asked.md | Lesson 002 content (report results with evidence) |
| contributors/fishtvlvoe/003-diagnose-first-system.md | Lesson 003 content (systematic debug flow) |
| contributors/fishtvlvoe/004-verify-before-speaking.md | Lesson 004 content (self-verify; don’t ask users to “try”) |
| contributors/fishtvlvoe/005-confirm-before-executing.md | Lesson 005 content (confirm path/branch/environment before ops) |
| contributors/fishtvlvoe/006-delegate-code-writing.md | Lesson 006 content (delegate code writing to other agents/tools) |
| contributors/fishtvlvoe/007-automate-repetitive-tasks.md | Lesson 007 content (automate repeated operations) |
| contributors/fishtvlvoe/008-label-assumptions.md | Lesson 008 content (label unverified assumptions) |
| contributors/fishtvlvoe/009-three-step-task-discipline.md | Lesson 009 content (align → record → execute) |
| contributors/fishtvlvoe/010-spectra-for-all-tasks.md | Lesson 010 content (require Spectra Change records) |
| _data/contributors.yml | Adds contributor metadata for discovery/filtering |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
| ## 操作前確認（強制） | ||
|
|
||
| 編輯/部署前告知： | ||
| - 檔案路徑：/Users/fishtv/Development/... |
There was a problem hiding this comment.
The example path includes a personal absolute filesystem location (/Users/fishtv/Development/...). For a reusable rule (and to avoid leaking local usernames/paths), replace this with a generic placeholder (e.g. <project_root>/... or /path/to/repo/...).
| - 檔案路徑：/Users/fishtv/Development/... | |
| - 檔案路徑：<project_root>/... |
| 分工： | ||
| - 業務邏輯/API/測試 → copilot -p "..." --yolo --model gpt-5.2 | ||
| - 需讀大量 codebase 再改 → kimi -p "..." --yolo --print -w <dir> | ||
| - 需跑 shell/測試驗證 → codex exec "..." |
There was a problem hiding this comment.
The suggested commands use --yolo (non-interactive / auto-approve style execution). Given these docs are intended as reusable workflow rules, please add a safety note or adjust the recommendation so destructive/high-impact changes still require review/confirmation before execution.
| 分工： | |
| - 業務邏輯/API/測試 → copilot -p "..." --yolo --model gpt-5.2 | |
| - 需讀大量 codebase 再改 → kimi -p "..." --yolo --print -w <dir> | |
| - 需跑 shell/測試驗證 → codex exec "..." | |
| 安全原則： | |
| - 先讓 Agent 產出計畫或 diff，再決定是否執行 | |
| - 任何會改檔、跑 shell、重構、批量修改的操作，禁止預設自動批准；必須先 review/確認 | |
| - 只有唯讀分析、低風險查詢可直接執行 | |
| 分工： | |
| - 業務邏輯/API/測試 → copilot -p "..." --model gpt-5.2 | |
| - 需讀大量 codebase 再改 → kimi -p "..." --print -w <dir> | |
| - 需跑 shell/測試驗證 → 先審查將執行的命令，再用 codex exec "..." |
| - [工具不通先找替代，不說「沒辦法」](001-fallback-before-giving-up.md)：工具失敗時的靜默降級策略 | ||
| - [任務完成主動報結果，不等被問](002-report-without-being-asked.md)：避免用戶焦慮等待 | ||
|
|
||
| ### Debug 與驗證 | ||
| - [先診斷再動手，不要猜](003-diagnose-first-system.md)：系統性 debug 流程，避免反覆修錯 | ||
| - [不確定的事自己查完再說](004-verify-before-speaking.md)：禁止把驗證工作推給用戶 | ||
| - [任何任務開始前確認路徑、branch、環境](005-confirm-before-executing.md)：三確認防止改錯地方 | ||
|
|
||
| ### 工作流設計 | ||
| - [超過 5 行代碼就外包給外部 Agent](006-delegate-code-writing.md)：主對話只做規劃和決策 | ||
| - [重複操作必須自動化，不手動重複](007-automate-repetitive-tasks.md)：任何需要改 N 次的事都寫 script | ||
| - [推測必須標註「還沒驗證」](008-label-assumptions.md)：分清推測和事實，避免誤導用戶 | ||
|
|
||
| ### Spectra SDD 流程 | ||
| - [每個任務強制三步：對齊→記錄→執行](009-three-step-task-discipline.md)：防止方向錯誤和無法追蹤 | ||
| - [所有開發任務必須建立 Spectra Change](010-spectra-for-all-tasks.md)：SDD Loop 無例外，不允許只存在對話紀錄 |
There was a problem hiding this comment.
This index uses relative links to the source .md files (e.g. (001-fallback-before-giving-up.md)). Other contributor pages use {{ site.baseurl }}{% link ... %} to generate correct URLs under a configured baseurl and to link to the built page instead of the raw markdown. Please switch these lesson links to the same {% link %} pattern (apply to all links in this list) so they don’t break on GitHub Pages / non-root deployments.
| - [工具不通先找替代，不說「沒辦法」](001-fallback-before-giving-up.md)：工具失敗時的靜默降級策略 | |
| - [任務完成主動報結果，不等被問](002-report-without-being-asked.md)：避免用戶焦慮等待 | |
| ### Debug 與驗證 | |
| - [先診斷再動手，不要猜](003-diagnose-first-system.md)：系統性 debug 流程，避免反覆修錯 | |
| - [不確定的事自己查完再說](004-verify-before-speaking.md)：禁止把驗證工作推給用戶 | |
| - [任何任務開始前確認路徑、branch、環境](005-confirm-before-executing.md)：三確認防止改錯地方 | |
| ### 工作流設計 | |
| - [超過 5 行代碼就外包給外部 Agent](006-delegate-code-writing.md)：主對話只做規劃和決策 | |
| - [重複操作必須自動化，不手動重複](007-automate-repetitive-tasks.md)：任何需要改 N 次的事都寫 script | |
| - [推測必須標註「還沒驗證」](008-label-assumptions.md)：分清推測和事實，避免誤導用戶 | |
| ### Spectra SDD 流程 | |
| - [每個任務強制三步：對齊→記錄→執行](009-three-step-task-discipline.md)：防止方向錯誤和無法追蹤 | |
| - [所有開發任務必須建立 Spectra Change](010-spectra-for-all-tasks.md)：SDD Loop 無例外，不允許只存在對話紀錄 | |
| - [工具不通先找替代，不說「沒辦法」]({{ site.baseurl }}{% link contributors/fishtvlvoe/001-fallback-before-giving-up.md %})：工具失敗時的靜默降級策略 | |
| - [任務完成主動報結果，不等被問]({{ site.baseurl }}{% link contributors/fishtvlvoe/002-report-without-being-asked.md %})：避免用戶焦慮等待 | |
| ### Debug 與驗證 | |
| - [先診斷再動手，不要猜]({{ site.baseurl }}{% link contributors/fishtvlvoe/003-diagnose-first-system.md %})：系統性 debug 流程，避免反覆修錯 | |
| - [不確定的事自己查完再說]({{ site.baseurl }}{% link contributors/fishtvlvoe/004-verify-before-speaking.md %})：禁止把驗證工作推給用戶 | |
| - [任何任務開始前確認路徑、branch、環境]({{ site.baseurl }}{% link contributors/fishtvlvoe/005-confirm-before-executing.md %})：三確認防止改錯地方 | |
| ### 工作流設計 | |
| - [超過 5 行代碼就外包給外部 Agent]({{ site.baseurl }}{% link contributors/fishtvlvoe/006-delegate-code-writing.md %})：主對話只做規劃和決策 | |
| - [重複操作必須自動化，不手動重複]({{ site.baseurl }}{% link contributors/fishtvlvoe/007-automate-repetitive-tasks.md %})：任何需要改 N 次的事都寫 script | |
| - [推測必須標註「還沒驗證」]({{ site.baseurl }}{% link contributors/fishtvlvoe/008-label-assumptions.md %})：分清推測和事實，避免誤導用戶 | |
| ### Spectra SDD 流程 | |
| - [每個任務強制三步：對齊→記錄→執行]({{ site.baseurl }}{% link contributors/fishtvlvoe/009-three-step-task-discipline.md %})：防止方向錯誤和無法追蹤 | |
| - [所有開發任務必須建立 Spectra Change]({{ site.baseurl }}{% link contributors/fishtvlvoe/010-spectra-for-all-tasks.md %})：SDD Loop 無例外，不允許只存在對話紀錄 |
| - username: fishtvlvoe | ||
| name: Fish（老魚） | ||
| role: 一人公司 / 獨立開發者 | ||
| work: 開發 BuyGo 電商管理系統（WordPress 外掛 SaaS） | ||
| stack: [WordPress, PHP, Next.js, Supabase, Claude Code, 多模型協作框架] |
There was a problem hiding this comment.
This new contributor is added to _data/contributors.yml, but contributors/index.md (the rendered contributors list/table) still only lists hanslin. Please add fishtvlvoe there as well so the new contributor is discoverable from the site navigation.
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 5ccc0f42e7
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| ## 我的經驗文章 | ||
|
|
||
| ### 多模型協作與路由 | ||
| - [工具不通先找替代，不說「沒辦法」](001-fallback-before-giving-up.md)：工具失敗時的靜默降級策略 |
There was a problem hiding this comment.
The contributor index links point to Markdown source filenames (e.g., 001-fallback-before-giving-up.md), but these lesson pages are rendered at permalink URLs like /contributors/fishtvlvoe/001-fallback-before-giving-up/ (see each lesson frontmatter), so GitHub Pages/Jekyll will not serve the .md path and users will hit 404s from this page. This makes the lesson list effectively unusable in production; switch these links to site URLs (e.g., {{ site.baseurl }}{% link ... %} as used in contributors/hanslin/index.md) or direct permalink paths.
Useful? React with 👍 / 👎.
新增 Fish（老魚）16 篇 Claude Code 踩坑經驗文章（016-031）， 涵蓋 CLI 優先（Supabase/Vercel）、CSS 驗證與漸層覆蓋、rsync 部署 SOP、 大檔案讀取效率、React hook 閉包陷阱、UI 截圖驗證、Next.js dev CORS、 iOS Safari HTTP 限制等主題。 https://claude.ai/code/session_019KgSecKkVXiZGbm91p4c4S
新增 fishtvlvoe（Fish 老魚）的 15 篇踩坑經驗文章，涵蓋工具可用性、 Bug 診斷 SOP、React hook 閉包陷阱、Vercel env 隱形換行、Postgres schema 演進等常見陷阱，建立貢獻者目錄並更新 contributors.yml 索引。 https://claude.ai/code/session_012nFPeqpWykpzT9K8e6s4Go
…ssons 合併遠端已有的 31 篇文章，新增本週 3 篇真正新增的踩坑經驗： - 032: 說「工具不能用」前先 which X 確認（L016） - 033: Vercel env add 禁用 echo，用 printf 避免隱形 \n（L051） - 034: Postgres schema 演進用 ALTER TABLE，CREATE TABLE IF NOT EXISTS 是 no-op（L052） contributors.yml 更新 lessons: 34 https://claude.ai/code/session_012nFPeqpWykpzT9K8e6s4Go
新增 Fish（老魚）的 11 篇踩坑經驗文章，涵蓋代理協作安全、 CLI 優先原則、React 閉包陷阱、Next.js 行動測試、 Postgres schema 演進、Vercel 環境變數等主題。
本週新增 2 篇踩坑經驗： - 035: 派工給外部代理前先 git commit（L050），防止 git clean 永久刪除 untracked 檔案 - 036: 中文路徑讓 Vite/Next.js Dev Server 白畫面（L061），@fs URL encode 問題需實體 mv 同步更新 contributors.yml：lessons 34 → 36
新增 6 篇本週踩坑經驗（037-042），涵蓋 L078/L081/L082/L079/L077/L076： - 037: 長任務 WIP commit 存檔點策略 - 038: Build 通過 ≠ 功能正確（行為驗證規則） - 039: 分析 UI 結構要讀 DOM 不要截圖 - 040: Cross-impact 分析（改 A 不壞 B 預檢） - 041: 程式碼風格衝突時必須明確選邊站 - 042: Agent tool call XML 參數值不能夾雜描述文字 跳過已覆蓋規則：L050（035 已有），L045（029 已有），L031（021 已有），L039（003-proactive-next-step 已有） https://claude.ai/code/session_01CPykLaw7EzrJmkvHjL4yr7
新增 2 篇尚未貢獻的通用踩坑經驗（L084/L090）： - 043-reuse-first.md：寫代碼前先搜尋現有實作，禁止重複造輪子 - 044-check-env-first.md：API key 先查 .env，不要問用戶 更新 index.md 新增「代碼效率與環境」分類，contributors.yml lessons 更新為 44 https://claude.ai/code/session_019g5iiMyv6NAm77aibPW1z5
…ost test L072: When switching dev server projects on the same localhost port, old service workers silently intercept requests and return stale cache. Symptom: curl works but browser shows wrong content with no console error.
新增 4 篇 Chrome MCP 進階操作踩坑文章（046-049）： - nativeInputValueSetter 填 React controlled input - find() ref 不能猜號碼、不能跨 batch 使用 - read_page 一次取所有 ref 再 batch 操作 - extension 權限錯誤直接換 tab 不要 debug 更新 contributors.yml：lessons 45 → 49 Co-Authored-By: Fish（老魚）<fish@fishot.com>
貢獻者
GitHub: fishtvlvoe（Fish / 老魚）
身份: 一人公司，開發 BuyGo 電商管理系統
新增內容
文章主題
多模型協作與路由
Debug 與驗證
工作流設計
Spectra SDD 流程
貢獻規則均來自 6 個月以上的 Claude MAX 日常使用踩坑紀錄。