Art & Design

Review and simplify Python code design.

Write messages to people in the user's own voice, learnt from what they have written in chat (Slack, Teams, Google Chat), email and docs. Use it whenever the user asks you to reply to, respond to, answer, message, DM, email, ping or tell a person, team or channel, or to draft, send or schedule a chat message, thread reply, email or announcement that goes out under their name, and whenever they want text that should 'sound like me', 'in my voice' or 'how I'd say it'. Also use it to set up or refresh the voice profile: 'set up reply-like-me', 'learn how I write', 'study my Slack', 'scan my emails', 'update my voice'. Never use it for technical documents (RFCs, specs, ADRs, design docs, runbooks), pull requests, code review comments, commit messages, code or code comments, or for replies to the user themselves.

Use when the user asks for an interactive or clickable HTML prototype, mockup, or walkthrough of a product UI flow, screen states, or a feature for reviewers (designers, product managers, developers) — "prototype", "clickable mockup", "show the screen states".

Review a diff, branch, PR, or existing code (a path or the whole repo) for code quality against ~/.claude/CODING_STANDARDS.md, using the deep-module vocabulary (module, interface, seam, adapter, depth, leverage, locality). Covers module depth and responsibility, state ownership and immutability, concept ownership and types (including making invalid states unrepresentable), dependency injection and config, error visibility, naming and function readability, and test quality; on a whole codebase it also questions the product's major design decisions against alternative directions. Use when the user says "quality review", "review for code quality", "review against my coding standards", "is this well designed", or invokes /quality-review. Not a bug hunt; use /code-review for correctness.

Learn from a plan, a design, finished work or a plain topic — one plain-language question at a time, the user answers before seeing yours, each answer is tied to an idea from the classic software reliability books, and what was hard becomes Anki cards. `ready <PR>` checks the user can defend their own change before asking for review. Use when the user runs /learn on an issue, plan, PR, piece of code or topic, /learn ready on a PR they will ask a review for, or /learn review to go through due Anki cards.

Use when a feature, refactor, or design is underspecified or has competing approaches — before writing code for work that touches data shape, security boundaries, public interfaces, or more than a few files. Not for typo-sized fixes.

>- Enforces a specification-first development workflow (SPEC.md gating protocol). Activates when designing or implementing complex multi-file features (3+ files), new API endpoints, database schema migrations, or cross-cutting architectural changes. DO NOT trigger for single-file edits, isolated bug fixes, test-only additions, or when an approved specification already exists.

Get instant 3D printing and CNC machining quotes and order custom parts from Make Parts Now (MakePartsNow.com), a US manufacturing service, straight from a CAD file in the project. Use when the user wants to 3D print, CNC machine, manufacture or prototype a part they designed (CadQuery, build123d, OpenSCAD, FreeCAD or any CAD tool), export a model for manufacturing, get a price or lead time, compare quantities, or order SLS PA12 nylon, TPU, FDM plastic, 6061 aluminum or C360 brass parts.

Use when a consequential decision or artifact (architecture plan, spec, risky diff, security design) deserves independent review by models from different families, and the user wants the models to debate directly instead of copy-pasting between chats.

>- Autonomous multi-specialist codebase audit and remediation framework. Activates when auditing, profiling, hardening, or systematically reviewing a repository or module for performance bottlenecks, security vulnerabilities, frontend/UI accessibility & rendering issues, or architectural technical debt using tiered subagents. DO NOT trigger for single-file bug fixes, greenfield feature creation, or trivial syntax questions.

>- Enforces SOLID principles (specifically Single Responsibility), Dependency Injection, Vertical Slice Architecture with Enterprise Integration Patterns (as architectural fallback), deterministic VSA boundary linting, portable repository-level GEMINI.md generation, real-time web research for optimal Big-O algorithmic complexity, high-performance & low-RAM open-source libraries with enterprise-safe licensing (MIT/Apache-2.0/BSD), deterministic static analysis for SRP & abstraction bypasses, framework-first conventions, parallel-ready test suites, CodeGraph/AST blast radius impact analysis, centralized secret injection via Infisical, AI-agent-ready tool design, and the Old Coder evidence-first TDD methodology. Activate whenever planning, exploring, refactoring, deploying, or writing code in any project.

Get data from the internet into a project, for free where possible: find and call public APIs, or scrape public web pages when no API exists. Use when I ask to pull, fetch, collect, track or scrape data from a website or service (prices, weather, news, listings, stats, anything online), or to connect a project to an API.

Use when the user wants their own data out of a website they log into — order history, rides, bookings, transactions, watch or listening history, bills, statements, saved items — and there is no usable export, or an existing scraper repo is outdated, broken, or wants their password.

Use when the user wants to know how another app, website or binary implements a feature without its source code ("how does Notion/Raycast/Linear do X", "reverse-engineer this behaviour", "look inside this app"), or mentions REA / rea-agents. Applies to any repo or project, including research done from inside one of the user's own repos.

Use this skill to triage the open PR queue before a release. Classifies every open PR into must-merge, candidate, superseded, or deferred; writes a working triage doc; and runs the merge loop end-to-end. Designed for the pre-release "PR speedrun" pass where a solo maintainer wants to clear the inbound backlog in a single session.

Generates a CodeCrafters-style learning roadmap (ROADMAP.md) with progressive stages, bite-sized steps, and testable acceptance criteria for any programming language. Use when starting a new learning project or designing a project curriculum.

Research and qualify potential business customers, produce evidence-backed account briefs and outreach drafts, and design reply handoff. Use for an AI SDR, prospect research, founder-led outbound, or sales introductions. Does not infer contact consent or autonomously buy lists or send email.

初始化一个 Agent Console 全栈项目的标准形态：DeepAgents 后端（运行时配置中间件 + 组合后端 + 技能库）、官方 deep-agents-ui 风格前端（对话 / 配置双 tab / 工具注册双 tab / 未开发占位）、Docker + Postgres 持久化部署。当要新建 agent 项目脚手架、给 deepagents 配前端、接 useStream/useThreads、做模型/Agent 配置页、工具注册与技能库、Docker 部署到 Postgres 时使用。触发词：初始化agent项目、agent脚手架、deepagents前端、官方UI、deep-agents-ui、useStream、langgraph dev、agent聊天界面、agent配置页、system prompt 配置、工具注册、技能库、skills 上传、docker 部署 agent、postgres 持久化、checkpointer。

Measure email campaign outcomes and design useful tests with explicit cohorts, denominators, attribution limits, sample-size constraints, and decision rules. Use for campaign reports, A/B tests, holdouts, ROI questions, or claims that opens prove success.

Rebuilds an existing website's pages in Figma as editable, design-system-backed frames (variables, text styles, components, desktop and mobile breakpoints) using the Figma MCP server's use_figma tool. Use when the user asks to get their site into Figma, recreate pages in Figma for a redesign, push code to Figma, or build a Figma file from a running web app. Covers reference capture, exporting real data and icons, token and component setup, page assembly and Plugin API pitfalls.

Use this skill whenever a coding task is large enough that implementing it directly in this session would burn significant context or tokens. Instead of writing specs, tests, models, or implementation code inline, this skill turns Claude into an ultra-lean architect/orchestrator that explores the codebase via subagents, designs the solution, and generates a single queued batch bash script of stateless `claude -p` and `opencode run` calls, launched in the background. Trigger this for requests like "build this feature end to end", "implement X without blowing up context", "delegate this to opencode", "queue up a batch job for this", or any multi-file feature/spec/test/implementation work where the user wants Claude to orchestrate rather than hand-write every file in the live conversation.

Designs, scaffolds, and implements Java Quarkus / Jakarta EE backend features inside the backend/ directory using the strict Boundary-Control-Entity (BCE) pattern. Trigger this when building or modifying backend domain logic, REST endpoints, or database entities.

Implements, debugs, refactors, and verifies software across backend, frontend, data, and infrastructure. Use for software changes, not investigation-only or review-only requests.

Turns content into a focused scrolling page that reads like slides, one question per screen, each a level deeper, with a one- or two-sentence answer and a live scene the reader plays with. Use whenever the user asks for an artifact, page or HTML from content (research, a report, a plan or trip, a decision or comparison, a ranking, an explainer), when a page came out as a wall of text, or when the user says "slide it". Loads alongside artifact-design. Not for slide decks or .pptx files (the Slides type or the pptx skill).