Skip to main content
Run any Skill in Manus
with one click
Jia-Hong-Peng
GitHub creator profile

Jia-Hong-Peng

Repository-level view of 11 collected skills across 3 GitHub repositories.

skills collected
11
repositories
3
updated
2026-07-12
repository explorer

Repositories and representative skills

principal-engineer
software-developers

Use when Claude Code must implement changes or perform other nontrivial repository work with principal-level technical judgment: audit and improve a project, deliver a feature or bug fix, change legacy code safely, diagnose runtime behavior, make a technical decision or migration, or prove a change ready to land. Action-first: inspect real code, establish a baseline, implement the smallest safe change, run objective checks, inspect the diff, and stop explicitly. Excludes management, roadmap, process, and methodology advocacy.

2026-07-12
principal-engineer
software-developers

Use when Codex must implement changes or perform other nontrivial repository work with principal-level technical judgment: audit and improve a project, deliver a feature or bug fix, change legacy code safely, diagnose runtime behavior, make a technical decision or migration, or prove a change ready to land. Action-first: inspect real code, establish a baseline, implement the smallest safe change, run objective checks, inspect the diff, and stop explicitly. Excludes management, roadmap, process, and methodology advocacy.

2026-07-12
principal-engineer
software-developers

Use when Copilot must implement changes or perform other nontrivial repository work with principal-level technical judgment: audit and improve a project, deliver a feature or bug fix, change legacy code safely, diagnose runtime behavior, make a technical decision or migration, or prove a change ready to land. Action-first: inspect real code, establish a baseline, implement the smallest safe change, run objective checks, inspect the diff, and stop explicitly. Excludes management, roadmap, process, and methodology advocacy.

2026-07-12
rag-to-skill
computer-occupations-all-other

從 JSONL(RAG 源文件)建立 Claude Code Skill 的完整工作流程工具。 涵蓋「從零到 100% 完成度」的七個階段:JSONL 掃描 → 架構規劃 → 骨架生成 → 內容填充 → RAG 錨點標注 → 驗證稽核 → 最終交付。 本 skill 完全獨立,不依賴任何特定 repo 的 CI 腳本,可帶到任意專案使用。 何時觸發: - 使用者說「把這個 JSONL/RAG 轉成 skill」 - 使用者說「幫我從這本書建立一個 skill」 - 使用者有 .jsonl 檔案並想建立可分享的 Claude Code skill - 使用者說「建立新的 skill,來源是 XXX」 - 使用者說「我要把 paidRag 裡面的書做成技能」 適用對象:skill 開發者、知識庫維護者、想把書/文件轉成可執行 AI 知識的任何人。 典型提問: - 「把 紫微攻略.jsonl 做成 skill」 - 「幫我從 RAG 建一個 skill,100% 完成度」 - 「這本書的 JSONL 要怎麼變成可以給別人用的技能?」

2026-05-10
pdf-ocr
software-developers

PDF → JSONL,會先判斷是文字型還是掃描型: - 文字型:直接用 PyMuPDF 抽文字(不浪費 token) - 掃描型:用 LLM 視覺能力(Read / view_image)逐頁辨識 嚴禁任何傳統 OCR 引擎(Tesseract / PaddleOCR / EasyOCR)、雲端 OCR API、MCP 圖像工具。 一次只做一本書(單機鎖),完成後清除所有暫存(PNG 與 progress 檔)。 每次最多處理一個 batch(預設 8 頁),支援 --resume;長時間任務必須採用低上下文 append-only 流程。 觸發時機: - 使用者說「OCR 這個 PDF」 - 使用者說「把掃描 PDF 做成 JSONL」 - 使用者說「/pdf-ocr <path>」 - 使用者說「用 LLM 視覺辨識這個 PDF」 典型呼叫: - /pdf-ocr /path/to/book.pdf - /pdf-ocr /path/to/book.pdf /path/to/output.jsonl - /pdf-ocr /path/to/book.pdf --resume - /pdf-ocr /path/to/book.pdf --batch 8 --dpi 120

2026-05-10
Showing 3 of 3 repositories
All repositories loaded