Coding & Development

Orquestra o agy (Antigravity CLI, Gemini) como segundo agente do Claude Code — delega tarefas, roda trabalho em paralelo e pede revisão independente de código, escolhendo o modelo mais novo do agy na hora. Use sempre que houver agy instalado e a tarefa envolver gerar ou editar imagem (banner, foto, mockup, logo, texto dentro de imagem), transcrever ou analisar áudio/vídeo (.ogg, .mp3, .mp4, áudio de WhatsApp), dividir um trabalho grande em partes independentes, refactor mecânico em muitos arquivos, pesquisa longa, economizar cota do Claude, ou quando o usuário pedir "segunda opinião", "revisão cruzada", "valida esse código", "pede pro Gemini/agy", mesmo que não diga "agy".

Use when dispatching coding work through the gbdev CLI (controller as sole decider, Codex CLI and Cursor cursor-agent as local workers): first-use home/launcher init then webhook notify setup (Step 0), starting jobs, pipelines or runs, following progress via webhook wake-ups or polling, answering attention decisions, on-demand inspect evidence, landing, cleanup, and reminding the user how to watch the tmux live dashboard.

Assess any codebase — frontend, backend, mobile, data or infrastructure, any language — against the rule catalog in rules/, and produce client-ready HTML assessment reports (deep architecture review, executive decision brief, portfolio dashboard). Use when asked to assess, review, audit, or produce an assessment report for a codebase, or to regenerate a report from an existing assessment JSON.

The definitive master skill for building elite, production-grade full-stack web applications. Combines modern visual craftsmanship (interactive 3D WebGL/R3F canvases, custom GLSL shaders, kinetic typography, Lenis smooth scrolling, bespoke micro-interactions) with decoupled, fault-isolated backend architecture (Vertical Slice Architecture, single-file blast-radius containment, minimal 10-line idiomatic code, real senior developer comments). Provides multi-database mastery (PostgreSQL, Supabase, MySQL, SQLite), multi-gateway billing (Stripe, Lemon Squeezy, Paddle, PayPal), hardened enterprise auth (OAuth, Passkeys, RBAC, SAML), end-to-end testing & bug finding (Vitest, Playwright, k6 load testing), OWASP Top 10 anti-hacking security, multi-tier caching (Browser, CDN/Edge, Redis, DB), and scalable DevOps (Docker, Kubernetes). Use whenever the user requests building, architecting, optimizing, securing, or auditing a full-stack web application, 3D interactive site, or high-performance SaaS.

Use when creating, uploading, updating, publishing, or re-binding a local Agent Skill directory through Tracecat MCP, including pushing a fix to a skill an agent preset already uses.

Evaluate and negotiate SaaS and enterprise-software renewals, by zRev AI. Use when reviewing vendor quotes, order forms, contracts, pricing proposals, usage reports, email chains, screenshots, spreadsheets or meeting transcripts; separating real commercial savings from savings that only come from cutting scope; drafting counters, follow-ups, escalation emails, call plans or signature recommendations; and running a live renewal negotiation from first quote to signed order form. Works from uploade

Self-evolved security, reliability, and correctness guardrails enforced by Kintsugi AST/Regex scanner. Activate before editing code or running shell commands.

Use when someone wants the Claude desktop app signed in to several Claude accounts at once on a Mac, one app instance per account switchable from the Dock, or asks for separate Claude desktop profiles, a multi-account Claude setup, or Dock launchers for the Claude app.

Thiết lập và xây dựng plugin cho coding agent (Claude Code, Codex, Antigravity): hỏi rõ yêu cầu, sinh manifest cho từng host, skill, agent, command, hook, MCP/LSP server, README và marketplace, rồi tự kiểm tra bằng script và CLI chính thức (claude plugin validate, agy plugin validate). Dùng khi người dùng muốn tạo plugin mới, thêm component vào plugin có sẵn, chuyển plugin sang host khác, hoặc tạo marketplace để phát hành plugin lên GitHub. Ví dụ: "tạo plugin Codex", "đóng gói skill và MCP thành

Rehearse the trigger rules of a background or event-driven agent before you switch them on. Turns whatever the rules are written in (an agent hub's rules file, routines, CI "on:" blocks, cron lines, no-code automations) into one neutral rule sheet, lints it for self-triggering loops, unguarded chains, missing caps and double handling, then replays real or recorded events through it to show every run that would have started, what was merged, skipped, queued or dropped, and what it would cost per

Verifica una bibliografía (archivo .bib, lista de referencias de un paper/tesis, o el proyecto LaTeX que las cita) contra OpenAlex, Crossref, arXiv, OpenLibrary y NASA ADS; detecta citas falsas o inventadas (alucinadas), papers retractados, datos erróneos (año, volumen, página, DOI, orden de autores), duplicados, claves \cite sin entrada y preprints ya publicados; comprueba si cada paper respalda lo que el texto afirma al citarlo; sugiere referencias faltantes; escribe un .bib corregido (DOIs, e

Use when someone wants music recommendations or a Spotify playlist for a mood, activity, occasion, culture, or discovery request, wants to refine an earlier selection, or explicitly wants to manage their music taste memory or import listening history. Includes vague music requests and interview mode. Use the available Spotify tools for literal lookups and playback-state questions.

Vet an agent skill before you install it — stealing, spying, hidden sending, proven with file-and-line evidence. Static, read-only, pure Python.

Thiết kế và khởi tạo trọn bộ thư mục Agent Skill theo chuẩn kiến trúc 3 cấp độ Progressive Disclosure (SKILL.md, template.md/templates/, examples/, references/, scripts/) kết hợp tiêu chuẩn tự kiểm chứng và markdownlint, cho một hoặc nhiều provider (Codex, Claude Code, Antigravity). Ràng buộc bắt buộc: mọi lựa chọn đưa cho người dùng đều hỏi dưới dạng danh sách lựa chọn có khuyến nghị và dừng chờ trả lời; phân tích ý tưởng, đề xuất hướng tối ưu và hỏi lại mọi điểm chưa rõ thay vì tự quyết; hỏi p

卡在"同一个任务反复迭代仍不收敛"时的定向破局：这个 skill 不给更好的主意，换的是找做法的地方；另一半是审计已有的跨域对应（判结构，不判真值）。用于卡住了/死循环/僵局/死锁/破局/走不通/想不出下一步/deadlock/stuck/blocked；也用于自查。即使没人这么说，出现其一即触发：同一路子已退回两次；改了三轮，每轮的下一步都只是上一轮失败的直接后果、症状没消；同一目标上来回问答已 ≥3 轮仍未达成（用户反复问 AI，或 AI 反复问用户同一件事）；AI 本版回复的实质动作是让用户在候选里挑；AI 准备提的候选没有一条是用户没提过的。审计侧触发：手上已有一个跨域类比/对应关系，要判它能不能进推理链或搬回本域；或出现"这个类比结构对不对/自洽吗/据此能推出结论吗"，即使没人说"审计"二字。不用于无卡点的主题头脑风暴、下一步已知只是不想做、缺事实或只有用户知道的约束（那就问准）、判类比真假（审计只判结构，真值要领域证据）、润色解释。 English — unblocking when a task is iterated without converging; and au

Alur kerja langkah demi langkah untuk membuat skill AI agent baru dari ide atau kebutuhan nyata - menentukan apakah layak jadi skill, menulis SKILL.md yang valid, dan mengujinya sebelum dipakai. Gunakan saat user ingin membuat skill, menulis SKILL.md, memperbaiki skill yang tidak pernah aktif, atau bingung memilih antara skill, agent, command, dan hook.

Connect two Claude Code sessions running on different machines (same or different accounts) so the agents can message each other and work through a shared task, such as a backend agent with a mobile or web frontend agent running a test cycle together. One side runs `/agent-bridge host --role backend`, which starts an encrypted relay exposed through ngrok or the LAN and prints a pairing code. The other runs `/agent-bridge connect` with that address, the code and its own role (e.g. mobile). Use th

Voice every line of an animated short with the person's own Volcano Engine Seed-Audio key, keeping each character's voice consistent. Also generates ambience and sound effects. Use for character voices, dialogue audio, voice acting, ambience and sound effects.

Install, update, or check the hostel CLI on macOS, Linux, or Windows, and connect the computer to a Hostel server. Use it when the user says "install hostel", "update hostel", "hostel command not found", "connect to our Hostel server", "I have an invite code", "disconnect", or "open the Hostel app".

Architecture and integration rules for React web apps (Next.js App Router or React with Vite): server and client component boundaries, one API layer with a typed error model, state ownership, react-hook-form + Zod with server errors mapped to fields, state boundaries (loading.tsx, error.tsx, Suspense), i18n architecture (locale registry, resolution order, URL policy, request-scoped locale, catalogues, kit and date locale), RTL with logical properties, auth, XSS, CSRF, secrets and cache isolation

Understand and change Hostel platforms - the defaults (ios, android, macos, windows, linux), what each needs on the machine, devices and labels, and your own platforms with their scripts. Use it when the user says "add iOS/Android to a machine", "what does the Mac need for iOS", "Android emulator does not start", "add a device", "make my own platform" (for example tvOS), or "change the provision/prepare script".

用 Three.js 或 Blender 制作可交互的三维结构科普、机械拆解、装配演示和工厂工艺流程可视化。继承 Anatomy 的部件语义与状态驱动方法，输出真实三维几何、爆炸视图、部件说明和过程演示；用于讲清对象内部如何组成和工作，不用于制造级 CAD 或把示意动画称作物理仿真。

Set up, configure, back up, or update a Hostel server on a Linux VPS or a home machine. Use it when the user says "set up a Hostel server", "deploy hostel", "run hostel server", "put Hostel behind Caddy/a proxy", "change a server setting", "server.toml", "back up Hostel", or "update the server".

Add, change, or remove the machines that give Hostel its rooms - a static host over SSH, a machine behind NAT that joins the server, Namespace on-demand machines, or your own external provider. Use it when the user says "add a Mac", "add a Windows machine", "add my Linux box", "the machine is behind NAT", "use Namespace", "more rooms", "drain/remove/repair a machine", or "destroy a room".