skills

預設的系統化開發工作流 (SDD)。當用戶明確要求開發新功能、修復複雜問題、進行需求梳理、架構設計、任務規劃、程式碼實現、spec review / optimization，或在 project review / live demo 前盤點真實整合風險時使用。若工作同時需要處理 review rejection、retro finding、known issue、tech debt、change-request follow-up、或其他 continuous-improvement 請求，並且要判斷它應該 `continue active spec`、走 `CR against completed spec`、先進 issue log，還是真的需要 `new spec`，也應使用此 skill。若工作同時需要規劃需求到任務的主流程，並在 implementation / closeout 前安排 folder-level `TESTS.md` row-level 更新、workspace `.agents/specs/TESTS.md` reconciliation / rollup refresh、test evidence 回寫，或明確決定何時 handoff 給 `test-registry-manager`，也應使用此 skill。直接的 `TESTS.md` catalog cleanup / duplicate-ID / stale-row / mapping reconciliation 本身仍屬於 `test-registry-manager`。

把這個 skill 當成 spec 工作的前門與路由入口，並在需要 branch-spec authoring / resume / improvement classification 時指向本目錄的 `WORKFLOW.md`。當使用者要建立新 spec、續做做到一半的 spec、根據 `NEXT_STEPS.md` 詢問下一步、盤點或更新 `SPECS.md`、建立或整理 `RTM.md`，或遇到 review rejection、retro finding、tech debt、known issue、test gap、CR follow-up 等 continuous-improvement 請求而不確定應該併回既有 owner、走 CR overlay、先進 issue log，還是真的需要開新 spec 時使用。把 project-level system architecture / `.agents/steering/{product,tech,structure}` / architecture review / 架構 HTML 導向 `system-architect`，把 AI/security/privacy/PII/log/regulatory compliance inventory / internal-audit gap table 導向 `iso-ai-security-auditor`，把 `ISSUE_LOG.md` 治理導向 `issue-log-manager`，把 `TESTS.md` 治理導向 `test-registry-manager`，把 `SPECS.md` registry sync 導向 `spec-registry-manager`，並在真正需要 local dev / UAT / E2E runtime allocation 時轉交 `local-infra-registry-governance`。不要用在單純 folder-level `TESTS.md` 維護、單純 `SPECS.md` 更新、單純 compliance legal advice/certification verdict、或單純 local env 操作這些已明確屬於下游 skill 的情況。

負責掃描專案內的所有 Specs 文件，進行規格盤點與狀態更新，並生成或維護全局規格註冊表 (`SPECS.md`)。當使用者要求「盤點 spec」、「更新 SPECS.md」、「建立規格目錄」，或需要管理 completed spec 的 change request、cross-spec 影響、external contract 依賴治理、continuous-improvement fragmentation 風險摘要、以及跨 spec 的 live-demo readiness / false-green review 風險摘要時使用。這個 skill 不負責 live local infra runtime registry，也不直接重做 runtime 驗證。

管理 `TESTS.md` 與 test traceability 的專用 skill。當使用者要更新、盤點、刷新、reconcile、audit、clean up folder-level `TESTS.md` 或 workspace `.agents/specs/TESTS.md`，補 `Test ID`、`Owner`、`Canonical Command`、`Evidence Ref`、`Task / Spec Trace`、`Requirement / AC Trace`，處理 duplicate test IDs、stale rows、`unmapped_to_spec`、missing evidence，或在 spec closeout 前刷新測試治理時，都應優先使用此 skill，即使使用者沒有明說 skill 名稱。若工作同時涉及 open CR、review-pending baseline change、或需要重新判定 critical test evidence freshness，也應使用此 skill。不要用在新 spec authoring、`SPECS.md` registry sync、`RTM.md` authoring、最終 readiness verdict、或 local env / runtime work。

為 git 管控的專案建立跨 AI agent 的 hybrid bridge：repo-local `skills` 用 symlink / Junction 避免重複，`specs` 可選擇 sync 或 symlink，其餘 `.claude`、`.kiro`、`.codex` 內的設定 / 權限檔維持 real-directory + sync workflow。當使用者提到「設定 cross-agents symlinks」「初始化 agents 設定」「skills 不要重複」「保留 Claude/Codex 權限檔」「specs 要可切換 sync 或 symlink」「CLAUDE.md symlink」或要整理 cross-agent `.gitignore` 規則時使用此 skill。

Perform code review using Code Review System CLI tools. Use when agents need to analyze code, generate improvements, create reports, perform architecture analysis, inspect bounded context, query GraphRAG state, govern local GraphRAG artifacts, or produce generic producer-side routing handoff artifacts. Supports file-level and project-level reviews.

Map a code file and its symbols to a single business capability, optionally grounded in the project's declared requirements / business value, through either agent-instruction mode or explicit LLM-API mode. Used by Layer 5 capability enrichment; defaults to subscription coding-agent instructions and never calls provider APIs unless llm-api mode is explicitly selected.

Generate a concise file-level summary through either agent-instruction mode or explicit LLM-API mode. Used by summary enrichment; defaults to subscription coding-agent instructions and never calls provider APIs unless llm-api mode is explicitly selected.

Set up and run cross-agent / cross-model code review — where one coding agent's work is independently reviewed by a DIFFERENT model family to avoid single-model blind spots. Use this whenever the user wants a second-opinion / QA / QC review from another agent or model, wants to configure or switch which backend reviews (claude / codex / kiro / opencode / antigravity), wants to auto-trigger review when an agent finishes work (stop-hook), wants to run a one-off independent review of the current changes, or wants to update the SOTA model-card registry. Trigger this even when the user only says things like "have another model check this", "cross-check my changes", "review by a different agent", "set up automatic review when I stop", "install the review hook", or "which model should review this" — the whole point is independent cross-family review, so lean toward using it.

容器編排模式，涵蓋多實例服務部署、資料庫配置、基礎設施生命週期管理、KIND 叢集部署（Podman rootless）。當需要 (1) 部署多實例資料庫（如每個 app 獨立 Neo4j），(2) 在已獲得 registry-governed allocation 後配置服務資源，(3) 管理 Podman rootless 容器生命週期，(4) 設計共享開發基礎設施，(5) 設置 KIND 叢集並選擇 Podman/Docker 運行時，(6) 實作持久化儲存策略時使用。

Execute coding tasks from Scrum Master using TDD workflow and existing skills

Judge the quality and effectiveness of EXISTING tests, not just whether they pass. Use when reviewing or refactoring test code, when asked "are these tests any good / thorough / brittle", when tests pass but you suspect weak assertions, or as the test-review pass of a code review. Flags test smells (assertion roulette, mystery guest, fragile/conditional tests), FIRST violations, missing boundary/negative cases, and test-pyramid inversion. Orchestrated by the code-review skill alongside capability-mapper and code-summarizer.

Use this skill when you need to generate a plan from structured inputs, validate a plan, run a demo/UAT plan, or inspect a generated report through the packaged `uatdemo` CLI delivered as the `uat-demo-agent` skill bundle. Also use it when the user explicitly wants the installed workspace/global agent skill surface rather than direct package-level Go calls.

當使用者要把 `uat-demo` 與 SDD/spec 治理流程一般化到 target repo、需要 reusable governance templates、釐清 `spec-master` 與 `spec-driven-development` 邊界、或規範 bootstrap / UAT-to-demo handoff 時使用。這個 skill 提供 repo-owned 的 reusable guidance 與 templates，不取代 `spec-master`、`spec-driven-development`、`uat-demo-agent`、或 `local-infra-registry-governance`。

Multi-LLM collaborative decision-making through structured committee discussions inspired by Double Diamond, Delphi Method, and Lean Coffee

盤點與清理 Docker / Podman images 與本機 container runtime 空間的 global skill。當使用者想列出 docker/podman images、只保留每個 repository 最新 N 個 images、找出 untagged / dangling images、輸出可編輯的 UTF-8-SIG CSV 審核清單、依 CSV 安全刪除 image、盤點 Docker/KIND/Podman 空間、清理 Docker BuildKit build cache、清理 Docker unused volumes，或評估 KIND cluster 是否可釋放空間時使用。特別適合需要先 inventory/dry-run/確認、再保守清理的 container image 與 disk cleanup workflow。

Use this skill when the user asks to inventory, audit, gap-check, or prepare internal audit evidence for repository security, cybersec, privacy, PII/personal-data protection, logging, regulatory, and AI governance against CNS/ISO 27001, CNS/ISO 27002, CNS/ISO 42001, NIST AI RMF, NIST AI 600-1 GenAI Profile, EU AI Act, GDPR/ePrivacy, Taiwan PDPA/cybersecurity law, CCPA/CPRA, US state privacy laws, Colorado AI Act, NIS2, EU Cyber Resilience Act, Digital Services Act, Data Act, HIPAA, GLBA, COPPA, FERPA, FCRA, ISMS, AI management systems, AI security, AI compliance, AI risk, SoA, internal audit, or minimum security/AI compliance requirements. This skill helps collect organization documents plus repo evidence, distinguish code-implemented controls from externally operated controls, and produce an audit-ready gap table; it is not a certification or legal verdict.

為「潛在使用者 / 潛在客戶」打造 value-driven、行銷導向的 showcase / demo 站台內容（公開首頁、landing、產品 demo 站、範例租戶），含明確的 value proposition、Amazon Working-Backwards PR/FAQ、與 gstack CEO/founder viewpoint 觀點陳述，並輸出可交付的 showcase 檔案（markdown 草稿 + GENERATION_GUIDE.md 生成備忘 + 中英雙語同頁 HTML + 可列印 PDF 傳單 + 可寄 email 的 zip 打包）。當使用者說「強化 showcase」「showcase 要更行銷」「demo 站要 value-driven」「給潛在客戶看的內容」「公開首頁太工程味/太技術」「marketing copy」「landing page 文案」「把 demo 站寫得更有賣點」「value proposition」「價值主張」「working backwards / 未來新聞稿 / PR FAQ」「用 CEO 觀點陳述定位」「我們的範例站要能轉換」「showcase 要雙語 / 中英雙語 / bilingual」「列印成 PDF 傳單 / print friendly」「showcase 列印會失蹤 / 印出來跑版 / 背景不見」「加上 manual / review 連結按鈕」「打包 showcase / manual / review 成 zip 寄 email」，或任何把面向潛在使用者的公開展示頁從內部/工程口吻改寫成 benefit-led、可轉換的行銷內容時，都應啟動此 skill——即使使用者沒有說出「行銷」二字。注意 audience 區分：本 skill 服務「潛在使用者」(showcase)；若要產出給「高階管理者/operator/評估者」看的 review/簡報，改用 project-review-skill。

專業的B端供應鏈產品需求文件產生助手，深度理解採購、庫存、物流、供應商管理等核心業務場景。當使用者需要產生PRD文件、梳理產品需求、規劃功能模組、設計使用者體驗、進行需求評審時使用。適用於需要系統化、標準化、高品質的B端供應鏈PRD文件場景。

統一需求管理工作流編排器。自動化執行 A→B→C 完整流程：需求分析 (ba-analyst) → 會議記錄 (meeting-ba) → 商業需求文件 (brd-writer) → 技術規格 (ba-analyst spec) → 交付。適用於需要端到端需求管理、多skill協同、自動化階段推進的場景。

Use this skill whenever code changes touch authentication, authorization, sessions, secrets, untrusted input, API endpoints, AI/agent integrations, CI/CD, cloud infrastructure, dependency supply chain, or other sensitive trust boundaries. It routes the agent to deeper review lanes by workflow, app layer, language, platform, and zero-day / assume-breach posture while preserving the existing benchmark tooling.

Scan a source file for OWASP Top-10 security patterns and rank findings by risk so review effort goes to the highest-impact issues first. Use during security-conscious code review, when asked "is this secure / any vulnerabilities / injection risk / secrets in here", or as the security pass of a code review. Flags hardcoded secrets, SQL/command injection, XSS, weak crypto, insecure deserialization, disabled TLS verification, and path traversal, each mapped to OWASP/CWE. Orchestrated by the code-review skill; it is the concrete detector behind the skill's prose security guidance.

台灣政府預算與標案追蹤分析工具。當需要追蹤政府預算流向、分析標案金流、調查廠商關係網絡、建立利益流向視覺化時使用。特別適用於 NPO、公民團體進行政府監督、調查司法或特定議題領域的資金鏈結。

Use this internal workspace skill when maintainers need to package, verify, release, or install the deliverable `uat-demo-agent` skill bundle from the canonical workspace bundle to catalog-resolved global skill directories. Trigger when the user asks to release the skill, refresh bundled multi-arch binaries, install/update the global skill, or verify workspace/global bundle consistency.

B端供應鏈靜態互動HTML產生器。當使用者需要產生供應鏈管理系統的靜態互動原型、將PRD檔案轉化為可互動HTML、製作採購/庫存/物流等B端頁面原型時使用。支援AT-UI、Ant Design、Element Plus等多種UI風格，產生包含完整CRUD互動、表格單驗證、模擬資料的高品質HTML檔案。

搜集網路上關於受害者家屬相關的新聞、採訪、論述、人權團體活動等素材。當需要為生命權平等協會等非營利組織網站收集司法不公、受害者家屬權益、相關社會活動等主題資料時使用。支援首次完整搜集和每日增量更新，提供 Markdown 格式輸出含 Mermaid 時間線圖表、URL 引用和圖片連結。

當使用者要設計、評估、修正或推廣 Agentic Scrum / agent swarm self-upgrade / review trigger / retrospective trigger / global-skill governance 時使用。也用在 agentic Scrum 需要納入跨 spec system architecture drift、architecture review signal、或架構 ownership / handoff 問題時讀取這個 repo 內的正式治理 references，而不是依賴零散 side docs 的情況。

Backend architecture patterns, API design, database optimization, and server-side best practices for Node.js, Express, and Next.js API routes.

Provisional organization-level CIS for NAELT (National Association for Equal Life Treatment / 生命權平等協會). Provides mixed-mode brand guidance covering mission/values, stakeholder positioning, advocacy posture, service identity, and digital expression. When users mention "NAELT brand", "NAELT CIS", "生命權平等協會品牌", or request brand-aligned outputs for this victim-rights advocacy NGO. NOT a formal official brand manual; clearly labels CONFIRMED, DERIVED, PROVISIONAL, and UNAVAILABLE rules.

Detect code smells in a source file and name the specific refactoring move plus the test safety-net it requires first. Use when reviewing code for maintainability, when asked "how should I refactor this / what's wrong with this file / is this too complex", or as the refactoring pass of a code review. Flags long methods, long parameter lists, deep nesting, god files, duplicated blocks, magic numbers, and complex conditionals, each mapped to a named Fowler refactoring. Orchestrated by the code-review skill alongside test-quality-reviewer.

Automatically extract reusable patterns from Claude Code sessions and save them as learned skills for future use.

Instinct-based learning system that observes sessions via hooks, creates atomic instincts with confidence scoring, and evolves them into skills/commands/agents.

跨平台開發輔助腳本最佳實踐指南。當需要創建開發輔助工具、自動化腳本、CI/CD 腳本時使用，特別是需要在 Windows/macOS/Linux 跨平台運行的場景。包含 Python 腳本模板、架構偵測、彩色輸出、HTTP health check、命令執行、測試結果解析等實用模式。適用場景：(1) Docker 建置腳本，(2) E2E 測試腳本，(3) 部署自動化腳本，(4) 任何需要跨平台執行的開發工具腳本。

當使用者要現代化 **stored-procedure-heavy 的 legacy relational database**、SQL Server / MSSQL 系統、將資料庫內的商業邏輯拆回應用程式 / class / service / worker、規劃 SQL Server → PostgreSQL 遷移、分析大型 `.sql` dump、盤點 stored procedure 相依、決定哪些邏輯應留在 DB 與哪些應移到 application service / async worker / event handler 時使用。若使用者在問 legacy DB strangler、stored procedure 拆解、transaction boundary、schema + proc migration、或 DB-embedded business logic extraction，即使沒明說 modernization 也應觸發。不要用在單純 query tuning、單條 SQL 修正、或一般 PostgreSQL schema 最佳化這類非 modernization 任務。

Epic 拆解專家。將 Epic 拆解為可執行的 Tasks，遵循 INVEST/DEEP 原則，支援 2-4 並行 workers，包含 contract 定義和整合測試。調用 ba-analyst-skill 進行需求分析，確保 task 品質。

Formal evaluation framework for OpenCode sessions implementing eval-driven development (EDD) principles

整合的前端「設計 + 工程 + 視覺複核」domain expert。當使用者要設計前端 / 排版 / 視覺效果、實作 web UI（頁面、元件、landing、dashboard、artifact）、選用 shadcn / open-design / gstack / A2UI 產出並 render UI、或在 code review 時需要有人以前端專家身分審查 CSS/JS/HTML/JSX 與「畫面實際長怎樣」（含 Playwright 截圖、排版、階層、對比、無障礙、design-system/品牌一致性）時，都應使用此 skill——即使使用者只說「這個畫面很醜 / 很 AI / 排版怪怪的」「幫我把 UI 做好看」「這頁能不能過 design review」「幫我看看前端這段」。此 skill 是既有前端 / 設計 skills（frontend-design-skill、shadcn、brand-guideline-company、web-artifacts-builder、frontend-patterns、theme-factory、webapp-testing、uat-demo-agent）的 orchestrator 與 code-review 家族的前端 domain-expert sibling，不重造它們。也用在需要為 spec / code-review 補上「render 出來看一眼」的視覺複核 gate 時。

Go 效能瓶頸識別與 Rust 重構自動化工具。當需要優化 Go 專案效能、識別計算密集型熱點、將高負載迴圈轉換為 Rust、建立 CGO 橋接、支援 x86_64/ARM64 多架構部署、或在 Edge/Embedded 環境中達到極致效能時使用。適用於機器人控制、即時影像處理、路徑規劃、加密運算等高效能場景。

Idiomatic Go patterns, best practices, and conventions for building robust, efficient, and maintainable Go applications.

管理 workspace `ISSUE_LOG.md` 的專用 skill。當使用者要記錄尚未能安全歸入既有 owned spec / active lane 的改善項、整理 repeated issue、做 cluster / owner resolution、判斷 issue 應留在 issue log、fold 回既有 spec/CR、或提升為新 spec / shared skill change candidate 時使用。不要用在正式 spec authoring、`SPECS.md` registry sync、`RTM.md` bridge authoring、最終 readiness verdict、或 `TESTS.md` row-level 維護。