Coding & Development

Encontra no transcript os trechos de ~60-110s que valem um corte viral (tópico fechado, gancho forte, payoff), valida as fronteiras de frase e apresenta ao usuário para escolha. Etapa 3 do Lumiere.

Sets up and operates PricePirate, the repricing app for Shopify stores selling on Idealo and Google Shopping, through the pricepirate-shopify MCP server. Use when the user wants to import Shopify products into PricePirate, connect them to marketplace listings, design or change pricing strategies, review or apply price changes, or analyse competitors and marketplace positions.

Universal Intent-to-Reality Master Engine - turns a high-level goal into a verified, running system with a minimal token footprint. No conversational padding; maps intent to constraints and AST schemas; scans dependency graphs, imports and downstream callers before changing code; delegates heavy auxiliary checks to sub-agents. All generated code must pass mechanical gates (compilers, tree-sitter, pytest, cargo test, node, csc, stdio JSON-RPC) in a read-trace, diagnose, patch, re-verify loop with at most 5 retries. Blends Python, C/C++, Java, C#, Dart, MicroPython, IPC sockets, BM25/vector search and state machines. Workflow is DISCOVER, PLAN & ISOLATE, EXECUTE, HEAL, DELIVER. Use when the user gives a high-level goal to build end to end, asks for a multi-domain system or pipeline, or says intent-to-reality, master engine, niyetten gerçeğe, uçtan uca kur.

Stage 7 of a wp-site-kit WordPress build - staging behind Cloudflare Access, server hardening, safe deploys from a commit (scripts/deploy-safe.sh), database content changes, the launch audit, DNS cutover without breaking a running business, and the first week after launch. Use for "deploy", "set up staging", "go live", "launch", "cutover", "move DNS", "harden the server".

Write tests that catch real bugs and survive refactors. Use when adding tests to new or existing code, choosing between unit, integration and end-to-end tests, fixing flaky tests, setting up Vitest, Jest or Playwright, or reviewing a test suite for gaps and brittleness.

Build Horizon native mods (.hnmod) — Rust mixin bundles compiled into a custom Horizon client binary. Use when the user wants code-level behavior in the Horizon game (camera, input, rendering, physics, gameplay hooks), mentions horizon-mixins, mixin injection, native hooks, NATIVE_HOOKS_API, .hnmod, native-mod staging, or porting a Java mod's mechanics to Horizon. Also use when debugging a native mod that fails to stage, build, or pin correctly. Even if the user just says "Horizon mod that changes how X works", this skill applies — content mods (horizon-content-mod skill) cannot run code.

Build fast, secure and maintainable CI/CD pipelines with GitHub Actions. Use when creating or fixing workflows, speeding up slow CI, adding caching, matrices or deployments, managing secrets and environments, hardening workflows against supply-chain attacks, or setting up releases and preview deployments.

Creates or upgrades a repository's AGENTS.md (read by Codex, Cursor, and other agents) and a CLAUDE.md that imports it, so Claude Code, Codex, Cursor and Gemini CLI share one set of project rules. Scans the repo to detect languages, package manager, real build/test/lint commands, framework, directory layout and existing rule files (.cursorrules, .cursor/rules, CLAUDE.md, copilot-instructions), then writes a concise, verified rules file. Use when the user asks to set up AGENTS.md or CLAUDE.md, "project rules", "make Claude Code and Codex follow the same conventions", onboard an agent to a repo, or migrate .cursorrules.

Stage 3 of a wp-site-kit WordPress build - local DDEV site, the core must-use plugin (modules), the presentation-only child theme with skins, the content model (custom post types, meta, WP_Query, render functions), and the enquiry pipeline that saves leads before forwarding. Use when building or changing features, templates, content types, forms or plugin modules on a wp-site-kit site.

Build, validate, and install Horizon content mods (.hmod) — data-driven mods for the Horizon game (the Rust rewrite of Minecraft Java Edition 26.3). Use when the user wants to create a Horizon mod, add recipes, blocks, items, loot tables, tags, or lang files, fix a mod that silently does not appear in-game, package or check a .hmod, or asks about horizon-mod.toml, capabilities, or the Horizon mod pipeline. Also use for porting Java datapack-style content to Horizon. Even if the user never says "content mod", any "make me a Horizon mod that adds X" request should trigger this skill.

Make or change a browser game with SceNoCo — three.js scenes in TSX (no React) and Unity's behaviours, from the @scenoco-three npm packages. Use when asked to build a 2D or 3D web game with SceNoCo, three.js and Unity-style behaviours, to add a game to a React app, or to test or debug a SceNoCo game in the browser through window.__scenoco.

Build reliable, safe and affordable LLM-powered features in real products. Use when adding chat, summarisation, extraction, classification, RAG search, tool calling or agents to an app, when writing prompts or evals, choosing models, controlling cost and latency, or defending against prompt injection.

Build, refactor and review Shopify Online Store 2.0 themes. Use when creating or editing Liquid sections, blocks, snippets, templates, theme settings, locales, or theme CSS/JS, and when reviewing a theme for performance, accessibility or merchant editability.

Use when building a feature or fixing a bug and you want a disciplined, minimal workflow - domain modelling (DDD), Gherkin scenarios as tests (BDD), smallest complete change, verified before claiming done. Triggers on "build", "add feature", "implement", "fix", or /sdd. Also use when continuing or planning work from an existing spec or plan (docs/specs/, docs/plans/), e.g. "continue with phase 3", "next step", "implement the spec".

Resolve git merge, rebase, cherry-pick, revert or stash-pop conflicts correctly - understand what each side was trying to do, combine them rather than pick one, check the result builds, and continue the operation. Use this whenever the user hits a merge conflict, sees `<<<<<<<` markers, "CONFLICT (content)", "both modified", "deleted by us/them", or asks to "fix the conflicts", "finish the rebase" or "resolve and continue". Also use it before running `--ours`, `--theirs`, `-X ours` or `-X theirs`, because during a rebase those mean the opposite of what they appear to, and taking one side wholesale silently discards someone's work.

Rewrite git history without losing work or clobbering other people - amend, reset, squash, reorder, drop or edit commits, change a commit message, remove a commit from a branch, or force-push. Use this whenever the user asks to amend, squash, fixup, "undo the last commit", "get rid of that commit", "remove a commit from history", "clean up my commits before the PR", or to force-push; and whenever a task would end in `git push --force`. It decides whether rewriting is even appropriate (a commit others have pulled should be reverted, not erased), takes a recovery point first, avoids the interactive editors that hang in a non-interactive shell, and pushes with a lease instead of a bare force.

Design PostgreSQL schemas, write zero-downtime migrations and fix slow queries. Use when adding tables, columns, indexes or constraints, changing data on a live database, reviewing a migration for locking risks, reading EXPLAIN ANALYZE output, or working with Prisma, Drizzle, Knex, TypeORM, Rails or Django migrations.

通过 MCP 自主操作 Unity 与 Blender，完成 3D 游戏资产的搜索复用、建模、清洗、导出、导入与验收全流程。核心原则是素材优先：优先从 MCP 内置素材源（Sketchfab、Poly Haven）与外部素材库（3d66 等国内站、TurboSquid、CGTrader、Kenney、Quaternius、Mixamo、AI 生成等）搜索复用现成资产，搜不到再在 Blender 自建。涵盖 MCP 工具发现与连接、bpy 自动化建模与批处理、素材审计清洗、Unity 正确 FBX 导出、Unity 导入设置（材质/LOD/碰撞/骨骼动画/Prefab）、批量导入后处理与故障排查。当用户要求用 MCP 操作 Blender 或 Unity、找 3D 模型/素材/动画、把下载素材导入 Unity、处理素材格式与授权、Blender 建模或改模、导出 FBX、搭建 Blender→Unity 管线、修复粉色材质/缩放/法线/轴/碰撞/动画异常时使用。 English: Asset-first 3D pipeline driving Unity and Blender through MCP: search and reuse existing assets, audit and clean them, export FBX, and import into Unity with correct materials, LODs, colliders, skinned animation and prefabs. Use when sourcing 3D assets, building a Blender-to-Unity pipeline, or fixing pink materials, scale, normals, axes, colliders or animation issues.

3A 级 3D 游戏开发文档（GDD）设计与审校技能：把模糊的游戏想法转化为可直接交付开发、达到 Steam 上架标准、对标 3A 大作的完整文档体系。覆盖游戏愿景与 Pitch、美术圣经、场景与关卡描述、主线玩法与系统数值、剧情叙事与台词脚本、交互与 UI/UX、角色设定、物品/道具建模规格、动画与帧动画规格、音频设计、技术架构与性能预算、QA 与 Steam 上架验收。当用户要写游戏策划案、GDD、游戏设计文档、场景/关卡/玩法/台词/交互/建模/动画规格、美术需求单，或需要审校游戏文档是否达到可开发、可上架、对标 3A 的质量时使用。支持 Unity/Unreal/Godot，附带一键脚手架脚本生成整套文档树。 English: AAA-grade game design document (GDD) authoring and review for Steam-ready games. Covers vision/pitch, art bible, level specs, gameplay systems and numbers, narrative and dialogue, UI/UX, character/prop/animation/audio specs, tech performance budgets, and QA plus Steam release checklists. Supports Unity, Unreal and Godot. Use when writing or reviewing a GDD, game design doc, level/scene spec, art brief, or release QA checklist.

Do the complete initial project setup for a fresh repository — interview the user, create CLAUDE.md, AGENTS.md, and project docs (PRD, ROADMAP, CONTEXT, STATE, SETUP, ADRs), install only the skills relevant to the project's stack, add a reviewer agent, and fix .gitignore — then verify, commit, and stop. Use when the user says "set up a new project", "bootstrap this repo", "initial project setup", "scaffold this project", or asks to set up a fresh repo for AI coding agents. Do NOT use for adding a feature, fixing a bug, or continuing work on a project that is already set up.

Write fast Mojo data code and find out where its time goes, from the per-row loops of a pipeline down to the readers, writers and codecs under it, and take benchmark numbers that survive being checked. Use when writing or reviewing Mojo code that loops over Arrow record batches, Parquet or Iceberg scans (arrow-mlake, parquet.mojo, iceberg.mojo, dista, orca), or that filters, groups, joins or aggregates rows; when making a parser, reader, writer, encoder or codec faster; when writing a benchmark or comparing against pyarrow, parquet-cpp, arrow-rs or Ray; when a number looks too good or moves between runs; or before claiming a speedup. Pairs with Modular's mojo-syntax skill.

让 AI 通过 MCP（或命令行）自主完成 Unity 3D 单机游戏的玩法逻辑开发：从 GDD 规格到可玩、可测试、可构建的代码。覆盖 Unity 6（6000.x，第一人称、URP、UGUI、新 Input System）下的项目脚手架与代码规范、输入系统、第一人称控制器、准星交互、Animator 状态机、UI 面板栈/HUD、背包/对话/存档/对象池/音频、NavMesh 敌人 AI、Profiler 性能定位、构建发布与测试。提供可直接编译运行的 C# 框架模板（带 asmdef 与测试）和一键脚手架脚本。当用户要求在 Unity 里写玩法代码、做控制器/移动/视角、交互系统、背包/对话/存档、UI 菜单、敌人 AI、音效集成、性能优化、跑测试或出包，或把 GDD 规格落地为可玩功能时使用。与 unity-blender-mcp（资产管线）、aaa-game-gdd（设计文档）配套，不负责 3D 资产制作、不覆盖多人联机。 English: Unity 6 single-player gameplay development from GDD specs to playable, testable, buildable code: project scaffold, Input System, first-person controller, interaction, UI panel stack, inventory, dialogue, save system, object pooling, audio, NavMesh enemy AI, profiling, build and tests. Use when writing Unity gameplay code, controllers, interaction, UI, AI, or shipping a build.

Build a secure Nomba webhook endpoint that verifies the signature, deduplicates deliveries and acts correctly on each event (payment_success, payment_failed, payment_reversal, payout_success, payout_failed, payout_refund). Use this whenever the user wants to receive, handle, verify or debug Nomba webhooks or "notifications"/"callbacks" from Nomba, in Express, Next.js, NestJS, Fastify, Flask, FastAPI, Django, Laravel, Symfony or plain PHP, including when they only say "confirm the payment on my server", "listen for successful transfers" or paste a payload with event_type and nomba-signature headers. Use it also when reviewing an existing webhook handler.

Split working-tree changes into atomic, single-purpose git commits - survey what changed, group it into logical units, order them so each commit builds, and write messages that match the repository's existing style. Use this whenever the user asks to commit, especially when the tree holds more than one logical change, and whenever they ask to split, separate, reorder, restructure or clean up commits. Also use it when a commit message would need the word "and" to be accurate, or when only part of a file should be staged - the interactive `git add -p` and `git add -i` silently stage nothing in non-interactive environments, so partial staging needs the techniques here.