zeuikli/claude-code-workspace / .claude/skills/research-best-practices
research-best-practices
Verified研究外部 Claude Code best practice 來源(GitHub repos、官方 docs、blog),對比現有 workspace 做 gap analysis,產出可操作建議清單。
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 4.6 KB |
| references/sources.md | markdown | 10.5 KB |
references/sources.md
View on GitHub# 預設研究來源 | Default Research Sources
> 最後更新:2026-04-18 — 本輪研究已驗證所有連結可存取
## 官方一手來源(Primary Sources)
| 來源 | 說明 |
|------|------|
| https://code.claude.com/docs/en/hooks | 完整 Hook 事件清單(25 種,含 exit code 規約)|
| https://code.claude.com/docs/en/best-practices | 官方最佳實踐總覽 |
| https://code.claude.com/docs/en/memory | Auto Memory + CLAUDE.md 結構 |
| https://code.claude.com/docs/en/routines | Routines(排程 / API / GitHub webhook)|
| https://code.claude.com/docs/en/mcp | MCP 整合指南 |
| https://code.claude.com/docs/en/skills | Skills 官方文件(格式 / 類型 / 設計原則)|
| https://code.claude.com/docs/en/permissions | 權限系統 + Wildcard 語法 |
| https://code.claude.com/docs/en/sandbox | Sandboxing 開源 runtime |
| https://code.claude.com/docs/en/statusline | Status Line 自訂 |
| https://code.claude.com/docs/en/settings | 37 settings + 84 環境變數完整清單 |
| https://claude.com/blog | Anthropic 官方 blog(最新功能公告)|
## Boris Cherny (@bcherny) — 全部已驗證文件(時序)
> Boris Cherny 是 Claude Code 創始人(Anthropic)。所有文件已直接 fetch 驗證。
| 日期 | 檔案 | 核心洞察 | Workspace 狀態 |
|------|------|----------|----------------|
| 2026-01-03 | [claude-boris-13-tips-03-jan-26.md](https://github.com/shanraisshan/claude-code-best-practice/blob/main/tips/claude-boris-13-tips-03-jan-26.md) | 5 個平行 Claude、Plan Mode、Tag @claude on PRs、PostToolUse 自動格式化、MCP(Slack/BigQuery/Sentry)、背景 agent 驗證 | ⚠️ 部分採納 |
| 2026-02-01 | [claude-boris-10-tips-01-feb-26.md](https://github.com/shanraisshan/claude-code-best-practice/blob/main/tips/claude-boris-10-tips-01-feb-26.md) | Git worktrees 平行、"Knowing everything you know now, scrap this" prompt、Ghostty terminal、"use subagents" 指令、資料分析 | ⚠️ 部分採納 |
| 2026-02-12 | [claude-boris-12-tips-12-feb-26.md](https://github.com/shanraisshan/claude-code-best-practice/blob/main/tips/claude-boris-12-tips-12-feb-26.md) | Wildcard 權限(`Bash(bun run *)`、`Edit(/docs/**)`)、Sandboxing(`/sandbox`)、Plugin marketplace(`/plugin`)、Output styles(Explanatory/Learning)、37 settings | ⚠️ 部分採納 |
| 2026-03-10 | [claude-boris-2-tips-10-mar-26.md](https://github.com/shanraisshan/claude-code-best-practice/blob/main/tips/claude-boris-2-tips-10-mar-26.md) | Code Review(多 agent PR 審查)、Test Time Compute(多個不相關 context windows)| ✅ 概念在 deep-review |
| 2026-03-25 | [claude-boris-2-tips-25-mar-26.md](https://github.com/shanraisshan/claude-code-best-practice/blob/main/tips/claude-boris-2-tips-25-mar-26.md) | Always squash merge、PR size p50=118 lines / p90=498 lines | 📝 參考 |
| 2026-03-30 | [claude-boris-15-tips-30-mar-26.md](https://github.com/shanraisshan/claude-code-best-practice/blob/main/tips/claude-boris-15-tips-30-mar-26.md) | /branch、/btw、/loop、/batch、--bare、WorktreeCreate hook、--add-dir | ✅ **已採納** |
| 2026-04-16 | [claude-boris-6-tips-16-apr-26.md](https://github.com/shanraisshan/claude-code-best-practice/blob/main/tips/claude-boris-6-tips-16-apr-26.md) | Auto Mode(Shift+Tab)、/fewer-permission-prompts、Recaps、/focus、Effort 五層、/go 驗證模式 | ✅ **已採納** |
## Thariq (@trq212) — 全部已驗證文件(時序)
> Thariq 是 Anthropic Claude Code 核心團隊成員。所有文件已直接 fetch 驗證。
| 日期 | 檔案 | 核心洞察 | Workspace 狀態 |
|------|------|----------|----------------|
| 2026-03-17 | [claude-thariq-tips-17-mar-26.md](https://github.com/shanraisshan/claude-code-best-practice/blob/main/tips/claude-thariq-tips-17-mar-26.md) | 9 種 Skill 類型、9 個 Skill 設計 Tips(T1-T9):description=trigger、${CLAUDE_PLUGIN_DATA}、On-demand Hooks、Progressive Disclosure | ✅ **已採納**(add-skill/SKILL.md)|
| 2026-04-16 | [claude-thariq-tips-16-apr-26.md](https://github.com/shanraisshan/claude-code-best-practice/blob/main/tips/claude-thariq-tips-16-apr-26.md) | Session 管理深度指南:Context rot ~300-400k tokens、Context carry-forward 光譜、Rewind > 口頭修正、Bad compact 成因、Subagent mental test | ✅ **已採納**(session-management.md)|
## Boris Cherny 完整 Tips 對照表(已採納項目)
| Tips | 重點 | 對應 workspace 位置 |
|------|------|-------------------|
| `/branch` — session forking | `claude --resume <id> --fork-session` 探索不同方向 | session-management.md |
| `/btw` — side query | 非阻塞式提問,不中斷 agent 進度 | session-management.md |
| `/focus` — focus mode | 隱藏中間工具輸出,只看最終結果 | session-management.md |
| Recaps | 回到 session 時自動顯示摘要 | session-management.md |
| `/loop` — recurring tasks | `/loop 5m /babysit` 定期自動執行 | routines.md |
| `/batch` — parallel migration | fan-out 到多個 worktree agent | routines.md |
| `--bare` flag | 非互動 SDK 用途,啟動速度最高 10x | routines.md |
| `WorktreeCreate` hook | 為 worktree 初始化非 git VCS 環境 | auto-sync.md |
| `--add-dir` / `/add-dir` | 多 repo 同時可見,跨目錄授權 | cli-enhancers.md |
| PostToolUse auto-format | `bun run format` 在 Edit/Write 後自動執行 | auto-sync.md(已記錄)|
| Auto Mode(Shift+Tab) | classifier 自動批准安全命令 | opus-best-practices.md |
## Thariq 完整 Skill 設計原則(T1–T9)
| 原則 | 說明 | Workspace 位置 |
|------|------|----------------|
| T1 | 別說廢話 — 只寫 Claude 不知道的事 | add-skill/SKILL.md(反模式)|
| T2 | Gotchas 是最高信噪比的內容 — 從失敗案例持續更新 | add-skill/SKILL.md(驗證清單)|
| T3 | Skill 是資料夾不只是 markdown — Progressive Disclosure | add-skill/SKILL.md(Step 3)|
| T4 | 別硬規定步驟 — 給目標與限制,讓 Claude 靈活 | add-skill/SKILL.md(反模式)|
| T5 | 用 `config.json` 存設定 — 未設定時詢問使用者 | add-skill/SKILL.md(Step 3)|
| T6 | description 是給模型看的 — 寫觸發條件不是功能摘要 | add-skill/SKILL.md(描述規則)|
| T7 | `${CLAUDE_PLUGIN_DATA}` — skill 跨 session 資料的穩定路徑 | add-skill/SKILL.md(Step 3)|
| T8 | 在 scripts/ 放腳本讓 Claude 組合,不從頭寫 boilerplate | add-skill/SKILL.md(Step 3)|
| T9 | On-demand Hooks — skill 啟動時才生效,session 結束即移除 | add-skill/SKILL.md(On-demand Hooks)|
## 社群驗證來源(Community)
| Repo | 重點內容 | 驗證日期 |
|------|----------|---------|
| [shanraisshan/claude-code-best-practice](https://github.com/shanraisshan/claude-code-best-practice) | Boris + Thariq tips(2026-01 至 2026-04)| 2026-04-18 |
| [howborisusesclaudecode.com](https://howborisusesclaudecode.com) | 粉絲整理 87 個 Boris tips | 2026-04-18 |
| [hesreallyhim/awesome-claude-code](https://github.com/hesreallyhim/awesome-claude-code) | 75+ 社群 repo 策展清單 | 未直接驗證 |
| [ykdojo/claude-code-tips](https://github.com/ykdojo/claude-code-tips) | Minimal CLAUDE.md 原則 | 未直接驗證 |
## 新功能(Research Preview + 正式功能,2026-Q1/Q2)
| 功能 | 版本 | 說明 | Workspace 狀態 |
|------|------|------|----------------|
| `/ultraplan` | v2.1.91+ | 雲端規劃:Claude 在雲端分析需求 → web review UI(inline comments / emoji / threads)→ teleport 回 terminal 執行 | ✅ 已加入 CLAUDE.md |
| `/ultrareview` | v2.1.86+ | 多 agent fleet 深度 code review;5–10 分鐘;Pro/Max 3 次免費,之後 $5–20/run | ✅ 已加入 deep-review/SKILL.md |
| **Remote Control** | v2.1.51+ | `claude remote-control` / `claude --remote-control` / `/remote-control` / `/rc`:從手機/瀏覽器控制本地 session;需 claude.ai 帳號 | ✅ 已加入 routines.md |
| **Channels** | — | `/en/channels`:Telegram / Discord / iMessage → Claude Code session;支援自訂 channel plugin | 📝 已加入 routines.md |
| Voice mode | Research Preview | `/voice` 或 `claude --voice`;支援即時語音對話控制 agent;Mobile app 優先 | 📝 參考,未加入 workspace |
> 來源:code.claude.com/docs/en/remote-control(直接 fetch 驗證);Thariq Twitter/X;Anthropic blog
## Thariq 工程哲學(2026 年公開洞察)
| 主題 | 核心論點 | Workspace 位置 |
|------|----------|----------------|
| **Unhobbling the Model** | 持續移除不必要的限制,讓 Claude 有更多自主性 | spec-interview/SKILL.md(核心理念)|
| **Delete-and-Rebuild Cycle** | 模型能力提升後積極刪除舊 scaffolding;10x 能力 = 10x 刪除 | spec-interview/SKILL.md(核心理念)|
| **Competitive Advantage as Vector** | 方向 + 累積洞察是護城河,而非單一功能 | 未採納(商業哲學)|
| **Seeing like an Agent** | 從模型視角設計工具;AskUserQuestion 3 次失敗→成功設計過程 | spec-interview/SKILL.md(架構哲學)|
| **Context rot ~300-400k tokens** | 注意力稀釋閾值;依任務類型有差異 | context-management.md、session-management.md |
## 驗證狀態更新(2026-04-18 深度審計)
| 項目 | 原狀態 | 驗證結果 | 備註 |
|------|--------|----------|------|
| "AutoDream" | 未驗證 | ❌ **確認幻覺** | 無任何官方文件 |
| "20%→50% skill activation" | 未驗證 | ⚠️ **社群來源** | community blog 有提及,非官方數字,勿引用 |
| "Prompt Caching Is Everything" by Thariq | 標題存在,內容未驗證 | ✅ **已驗證** | Thariq Shihpar,2026-02-24,3 個核心觀點(static-first、prefix matching、per-model caching)。內容已涵蓋於 context-management.md |
| Mobile `/teleport` | 未驗證 | ❌ **確認幻覺** | 官方文件無此指令;連接手機的正確方式是 `/remote-control` + QR code |
| `/remote-control` | 未驗證 | ✅ **已驗證** | 完整官方文件 code.claude.com/docs/en/remote-control;v2.1.51+;已加入 routines.md |
## 未驗證 / 疑似幻覺(仍不確定)
- "AutoDream" — ❌ 確認幻覺,無任何官方來源
- 具體效能數字("20%→50% skill activation")— ⚠️ 社群文章推測,非官方,不引用
- Mobile `/teleport` — ❌ 確認幻覺;正確為 `/remote-control` + QR code
## 下次審計建議查詢詞
```
site:github.com/shanraisshan/claude-code-best-practice tips
site:x.com bcherny claude
site:x.com trq212 claude
site:x.com bcherny claude code
site:x.com trq212 claude code
howborisusesclaudecode.com
site:github.com bcherny claude
site:github.com trq212 claude
claude.com/blog
code.claude.com/docs/en/
thariq.io
```
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| rea-changelog-updatemorluto/reaPrepare or rewrite REA release changelogs and GitHub release notes from pinned Git history, with verified contributor thanks and Release Please synchronization. | Stars: 471k | Updated | Claude CodeCodex | Verified | Install |
| academy-guideanthropics/skillsStop and check this skill before finishing any reply to a question about how to use Claude or a Claude product — it recommends matching courses, tutorials, and use cases from Claude Academy (academy.claude.com), Anthropic's learning hub. Trigger on: "how do I", "how can I", "getting started with", "what can Claude do", "teach me", "learn to use"; questions about artifacts, projects, skills, plugins, connectors, MCP; requests about rolling Claude out to a team, class, or organization; and any ask for training materials, onboarding content, or learning resources. Use it when the user is learning how to use a feature or product — not when they are mid-task and just want the task done. This skill composes with other skills: after consulting product documentation to answer how a Claude feature works, also check here for a matching course or tutorial — a docs-grounded answer and an Academy recommendation belong together. Only recommend on a strong match; never invent Academy content. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| claude-apianthropics/skillsReference for the Claude API / Anthropic SDK — model ids, pricing, params, streaming, tool use, MCP, agents, caching, token counting, model migration. TRIGGER — read BEFORE opening the target file; don't skip because it "looks like a one-liner" — whenever: the prompt names Claude/Anthropic in any form (Claude, Anthropic, Fable, Opus, Sonnet, Haiku, `anthropic`, `@anthropic-ai`, `claude-*`, `us.anthropic.*`, `[1m]`); the user asks about an LLM (pricing/model choice/limits/caching) — never answer from memory; OR the task is LLM-shaped with provider unstated (agent/MCP/tool-definition/multi-agent/RAG/LLM-judge/computer-use; generate/summarize/extract/classify/rewrite/converse over NL; debugging refusals/cutoffs/streaming/tool-calls/tokens). SKIP only when another provider is being worked on (overrides all triggers): OpenAI/GPT/Gemini/Llama/Mistral/Cohere/Ollama named in the query; OR `grep -rE 'openai|langchain_openai|google.generativeai|genai|mistralai|cohere|ollama'` over the project hits (run this grep FIRST if no provider named — don't Read the file). | Stars: 198k | Updated | Claude CodeCodex | Unverified | Install |
| doc-coauthoringanthropics/skillsGuide users through a structured workflow for co-authoring documentation. Use when user wants to write documentation, proposals, technical specs, decision docs, or similar structured content. This workflow helps users efficiently transfer context, refine content through iteration, and verify the doc works for readers. Trigger when user mentions writing docs, creating proposals, drafting specs, or similar documentation tasks. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |