skills

Audit, deduplicate, classify, normalize, convert, validate, and index game assets into a standalone, portable library suitable for Godot 4.x. Use when Codex needs to turn a messy downloaded asset dump into a clean asset library with purpose-revealing paths and filenames, preserve provenance and rights, identify corrupt or unsafe files, convert mixed image/audio/3D formats, create human-readable and machine-readable indexes plus a package-level AGENTS.md guide for AI discovery, or optionally check compatibility with an existing Godot project. The deliverable is an asset library, not a Godot project.

Design, audit, simulate, and operate game numerical systems and balance. Use for 数值设计, 数值策划, 平衡性, combat formulas, characters, skills, items and affixes, enemies and bosses, levels and XP, progression and content gates, currencies and sinks, rewards and loot tables, drop rates, pity and gacha, energy and time gates, difficulty and retention pacing, PvP matchmaking and ratings, meta balance, live-ops economies, telemetry, simulations, spreadsheets, configs, or balance patches. Turn a concept, GDD, table, config, telemetry set, or player complaint into experience targets, player models, formulas, parameter registries, content curves, scenario tests, instrumentation, and rollout criteria for RPG, MMO, action, shooter, roguelike, strategy, card, idle, casual, sports, racing, simulation, or hybrid games. Use game-remake-research first when reconstructing another game's undocumented systems from evidence.

Build, debug, and review modern React frontends with shadcn/ui as the default component system. Use when Codex needs to scaffold or extend a Vite + React + TypeScript (or Next.js) web app, install and compose shadcn/ui components, configure Tailwind and theming, wire forms, data tables, server/client state, dark mode, routing, or accessibility, or diagnose component, build, and browser runtime issues in a shadcn-based frontend.

Generate 2D game art assets — characters, monsters, NPCs, equipment, items, VFX, tiles, UI icons, and animation frame sequences — through OpenAI-compatible (gpt-image-2, gpt-image-1, dall-e) or Gemini image backends. Use when Codex needs to create or iterate on sprites and game-ready art with a configurable provider/base-url/model, transparent or chroma-key cutouts, and outputs prepared for downstream cutout (抠图), Spine skeletal animation, and paper-doll (纸娃娃) equipment-layering systems.

Build, debug, package, and review Tauri desktop GUI applications. Use when Codex needs to scaffold or modify a Tauri app, wire Rust commands and frontend calls, add Tauri plugins or capabilities, handle local persistence, platform integrations, signing, packaging, updater work, or release-readiness validation for a Tauri desktop app.

Generate or edit images with the Google Gemini API and Imagen models. Use when Codex needs to handle text-to-image generation, conversational image editing, multi-image grounding, background replacement, product shots, concept art, UI mockups, or model selection between Gemini native image models and Imagen 4 variants; run the bundled CLI (`scripts/gemini_image_gen.py`) and require `GEMINI_API_KEY` for live calls.

Format and review requirement documents, PRDs, meeting notes, or draft specs into a standard Markdown deliverable pack. Use when Codex needs to整理需求文档、仅整理原文格式并生成 `_formated.md` 文件，然后分别从产品经理、架构师、前后端研发视角识别业务、逻辑、交互、数据与实现问题，输出新的评审文档。

Phaser 3 browser game development, debugging, and feature-delivery skill for new or existing projects. Use when Codex needs to create or modify a Phaser 3 game, scaffold or extend a TypeScript + Vite setup, organize scenes, load assets, wire input, Arcade Physics, cameras, tweens, animations, or tilemaps, or diagnose browser runtime, asset-pipeline issues, or Vite and Rolldown build warnings in a Phaser repository.

Write, refine, and review Behavior-Driven Development specifications in Gherkin. Use when Codex needs to convert product requirements, user stories, acceptance criteria, workflows, or QA intent into Given-When-Then scenarios; identify business rules, examples, boundaries, and failure paths; or align product, engineering, and QA on executable behavior specifications.

Drive uninterrupted end-to-end software execution with minimal user interaction using the Ralph Loop engineering protocol. Use when the user wants Codex to keep developing continuously, avoid unnecessary questions, choose reasonable reversible assumptions, push through debugging and validation loops, recover cleanly from interruptions such as model-capacity errors, and only stop for clear completion or truly blocking issues such as destructive actions, missing credentials, legal or safety risk, or irrecoverable ambiguity. Trigger on requests like "don't ask me", "keep going", "work autonomously", "ralph loop", "ralph-loop", "闭环执行", "连续工作", "不要打断", "不要问我", or "一直开发".

Plan and execute multi-agent collaboration in Codex. Use when the user explicitly asks for subagents, delegation, or parallel agent work, and the environment exposes `spawn_agent`, `send_input`, `wait_agent`, and `close_agent`; or when an already-authorized task is large enough to benefit from splitting into independent explorer or worker subtasks. Coordinate one lead agent that keeps ownership of user communication, planning, integration, validation, and final decisions, while subagents take tightly scoped assignments under the same inherited model and configuration.

Source-backed analysis / 拆解 of an existing game into a remake / 复刻 teardown, benchmark / 对标, competitor study / 竞品分析, dossier, spec, formula set, experiment plan, handoff bundle, or vertical-slice research pack / 调研包. Use when the deliverable must separate original-game facts, inferences, and remake decisions. Do not use for generic brainstorming, casual recommendations, implementation-only work, or one-off mechanic explanations.