Coding & Development

Aspen Plus process-simulation assistant for chemical engineers. Builds models with the user stage by stage, writes input files (.inp), reads run files (.his, .rep, .cpm, .bkp) to troubleshoot runs and review models, checks balances and KPIs against expected values, runs, sweeps and edits copies of models by automation (COM), and answers how-to questions from the locally installed Aspen Plus help and example library. Use when the user mentions Aspen Plus or its files (.bkp .apw .apwz .inp .his .rep .cpm), blocks (RadFrac, RStoic, HeatX, Compr, FSplit), Calculator, Design Spec or Sensitivity, electrolytes, convergence or run messages, automating Aspen from Python, or wants to build or set up a model. Not for Aspen HYSYS, Aspen Adsorption or backup files of other programs.

Security-audit a large codebase or system with a swarm of agents: inventory its entry points (HTTP routes, GraphQL resolvers, webhooks, queue consumers, scheduled jobs, CLI commands, message handlers), split them into sections (fixed-size groups or themes), audit each section in its own git worktree with its own agent, collect every validated finding in the orchestrator worktree, then run a final chain audit that dedupes the findings and combines lower ones into higher-severity attack chains. Use when the scope is too big for one audit pass ("audit the whole backend", "split it up and audit each part", "security audit per section"). Options: `--by-group [N]` (groups of about N entry points, default 25) or `--by-theme`; asks when neither is given. Without a THREAT_MODEL.md it asks whether to build one first (recommended) or do a quick run. Keywords: sectioned audit, parallel security audit, vulnerability chaining, attack surface inventory.

Use when the user needs a local file converted between common image, audio, video, document, or data formats. Select installed tools, preserve originals, and verify the output.

Build, redesign, or restyle a desktop GUI app in PyQt6/PySide6 so it looks and feels like a polished macOS app (frameless window, traffic lights, sidebar, cards with soft shadows, animated micro-interactions, light/dark themes). Use whenever the user asks to make a desktop UI "красиво/стильно", "в стиле macOS/Mac", "современный интерфейс", "frameless window", "как в macOS", "сделай как в этом приложении" (referring to assets/app_template.py), or wants animations, blur/vibrancy, sidebar navigation, toggles, sliders, chat, drag&drop. Contains the exact build spec of the reference app app_template.py (tokens, per-widget geometry/timing/colors, interaction logic, page set) plus the must-know PyQt6-on-Windows crash pitfalls. Consult before writing any PyQt6/PySide6 UI code so the result doesn't look like a default Qt app.

Audit, fix and verify cookie, tracking and privacy compliance on any website (WordPress, WooCommerce, Shopify, Next.js, static, anything), for any target market - EU/EEA GDPR and ePrivacy, UK PECR, Switzerland, US state laws (CCPA/CPRA, GPC, wiretap/CIPA risk), Canada/Quebec, Brazil LGPD, Asia-Pacific and the rest of the world. Use this whenever the user mentions cookie banners, cookie consent, consent management tools (Complianz, CookieYes, Cookiebot, OneTrust, Usercentrics, iubenda, Termly and so on), GDPR/CCPA/LGPD or "privacy compliance" for a site, Google Consent Mode, Global Privacy Control, "Do Not Sell" links, trackers or pixels firing before consent, a banner that does not actually block scripts, privacy or cookie policy gaps, a legal or agency complaint about tracking, or asks whether a site is compliant, even if they only ask for one piece such as "add a cookie banner", "check what cookies this site sets" or "is GA4 legal here". Covers the whole job end to end - scoping jurisdictions, scanning the

Supply-chain and CI/CD security standards — npm/pnpm hygiene, the 14-item attack taxonomy (typosquatting, account takeover, cache poisoning, tag re-pointing, lockfile tampering, wormable malware), GitHub Actions hardening, Docker pinning, IOC quick-scan commands, and the per-project audit findings. Use for any npm/pnpm/lockfile work, any .github/workflows change, Dockerfile or image pinning, MCP/dev-tool config review, or a supply-chain audit.

Run the multi-model review loop on the current branch: Codex reviewers (Terra by default; Astra, Sol and Luna optional) and Claude reviewers (Opus by default; Fable, Sonnet and Haiku optional), with this session as the only fixer, through Code, Cleanup, Docs and Final phases. With no options it starts with terra:max:fast opus:xhigh code:20 spare:2. Typed options change the reviewers, each one's effort and Codex fast mode, and the review length, and `pick` opens a picker for them. Runs only when the user invokes /multimodel-review, because one run can use many max-effort reviewer runs.

Standardize raw, brownfield, or Terraformer-imported Terraform code (AWS, GCP, Azure) into production-ready HashiCorp standards: isolates provider.tf, versions.tf, and backend.tf, parameterizes hardcoded strings into typed variables.tf and locals.tf, replaces static cloud IDs with dynamic references, enforces ForceNew and zero-drift guardrails with moved blocks, and runs canonical HCL formatting, validation, and plan diff summaries.

Choosing and configuring three.js materials - unlit, diffuse, specular, PBR, toon, point and line materials - shared properties like blending and stencil, environment response, and the WebGPU node-material and TSL equivalents. Use when styling meshes or tuning material behavior.

Engineer production-quality Telegram bots, Mini Apps, and MTProto clients: message copy and /start flows, inline keyboards and navigation, handlers, callbacks and FSM state, webhooks, Telegram Stars payments and subscriptions, media and file processing, moderation, and initData auth, in aiogram, python-telegram-bot, grammY, Telegraf, or the raw Bot API, and Telethon or Pyrogram user clients and bot sessions (sessions, FloodWait, large files). Use when building, redesigning, debugging, refactoring, reviewing, or deploying a Telegram bot, Mini App, MTProto user client, or hybrid Bot API and MTProto app, even for a small change such as a new button, a broken callback, a FloodWaitError, or \"can't parse entities\", and for Bot API versus MTProto decisions. Russian: телеграм-бот, мини-апп, инлайн-кнопки, оплата звёздами. Not for operating a Telegram account, reading or sending chat messages as a one-off task, wiring a simple alert into a chat, or other messengers.

Creating and shaping three.js geometry: built-in shapes, 3D text, vertex buffers and attributes, merging, instancing, batching, point clouds and lines, on WebGL or with WebGPU and TSL. Use when building 3D shapes or text, custom vertex data, custom meshes, instanced meshes, or point clouds.

Act as a cost-aware head agent for complex tasks. Plan and decompose work, issue precise low-reasoning task packets to the cheapest capable subagents or deterministic tools, review worker reports, send targeted corrections, and integrate verified results. Use for multi-step projects or when efficient multi-agent delegation is requested; avoid delegation for trivial tasks.

운영 중인 네이티브 모바일 앱(iOS만, Android만, 또는 iOS·Android 두 앱)을 하나의 Flutter 앱으로 전환하는 전 과정을 단계별로 진행한다. 원본 코드 분석 → OS 간 비즈니스 차이 결정 → PRD → UI 인벤토리·디자인 명세 → 구현 지시서 → 공통 기반·모듈별 구현(서브에이전트 + 독립 리뷰어 승인) → 통합 테스트 → 실기기 QA 피드백 반영 → 서명·배포 설정 → 유지보수 문서까지, 각 단계의 문서를 직접 만들어 가며 진행한다. 사용자가 "네이티브 앱을 Flutter로 전환/마이그레이션/리라이트", "iOS·Android 앱을 Flutter 하나로 합치기", "Swift/Kotlin 앱을 Flutter로 옮기기", "크로스플랫폼으로 통합"을 말하거나, 네이티브 프로젝트 폴더를 주며 Flutter 전환 PRD·명세·작업 지시서를 요청하면 반드시 이 스킬을 쓴다. 전환 중간 단계(분석만, PRD만, 디자인 명세만, 이어서 구현, 구현된 Flutter 앱을 원본 UI에 맞추기)를 요청할 때도 쓴다. Also use for English requests such as "migrate/convert/rewrite a native iOS or Android app to Flutter", "merge our Swift and Kotlin apps into one Flutter codebase", or "write a Flutter migration PRD from these native projects".

Frontend unit and component tests with Vitest and React Testing Library: behavior-focused tests, accessible queries, user-event, network mocking with MSW, hooks, async UI and coverage. Use when writing or fixing tests for React components, hooks or pages, or when tests are flaky, slow or coupled to implementation details.

Build and debug game UI in Unity with UI Toolkit (UXML + USS + C# controllers) for any genre and platform - HUDs, health/ammo/stamina bars, interaction prompts, damage vignettes, menus, pause screens, score counters, cutscene screens. Covers the newer PanelRenderer workflow (no UIDocument) and classic UIDocument, writing a per-game UI style guide, layout without pixel guessing, and fixing UI Builder / USS / UXML problems. Use whenever the user mentions Unity UI, UI Toolkit, UXML, USS, PanelRenderer, UIDocument, UI Builder, a HUD, health bar, ammo counter or menu, or wants to restyle, fix or add any in-game interface in Unity, even if they never say "UI Toolkit".

Explain a pull request, diff, commit range or branch to a human reviewer — what changed, why, how data/control flows now, with a Mermaid diagram — then run a short interactive quiz to check they actually understood it before approving. Use whenever someone asks "explain this PR", "walk me through this diff", "help me review", "what does this change do", wants to onboard onto a teammate's change, or worries they're rubber-stamping AI-written code they don't understand.

Тестирование ПО (веб, мобайл, десктоп, бэкенд) по методике «Качество решает всё» Р. Суркова с дополнениями (OWASP API Top 10, Playwright, флейки, HTSM/HICCUPPS, SBTM, контракты, LLM-фичи) — требования, тест-дизайн, чек-листы, баг-репорты, UI, API, нефункциональное, автоматизация, TRR.

PostgreSQL with SQLAlchemy 2, covering schema design, indexes, query tuning with EXPLAIN, transactions and locking, Alembic migrations and zero-downtime schema changes. Use when designing tables, writing or reviewing queries and ORM code, a query is slow, or a migration must run on a live database.

Use when the user wants to see, preview, render, or screenshot a SwiftUI view or component without opening Xcode Canvas - checking layout, light/dark appearance, locale, fixture states, or local interaction of a macOS-compatible SwiftUI component. Produces PNG renders, an offline HTML theme gallery, or an interactive native macOS window from a fixture adapted from the project's real component.

Surface an agent session's whole delegation hierarchy as a LIVE, self-updating web page for a human who cannot see the terminal. A node server (serve.mjs) reads the session transcripts the agent harness already wrote, builds a spawn-tree graph of the root session and every subagent under it (type, task description, model, start/end time, alive or stopped, last log line), and serves ONE standalone HTML document at GET / with a GET /state JSON snapshot the page polls every 10s to patch itself in place; a switch on the page draws all agents (the default) or only the active ones. The TREE is served and the LOGS are fetched: a log arrives from GET /agent/<id> when the viewer selects that agent…

Help the user understand an existing software project without requiring them to read source code. The AI inspects the repository, explains business scenarios and architecture in plain language, and checks understanding only after teaching the necessary context. Use for project onboarding, business-flow learning, architecture understanding, and tracing how a feature works. Do not use for implementation or ordinary code review.

Edge-to-edge layout in Compose Multiplatform for Android and iOS - enabling it per platform, WindowInsets with Scaffold, lists, bottom bars, keyboards (IME), display cutouts, system bar icon contrast, and ad banners at screen edges. Use when building any screen scaffold, fixing content hidden behind the status bar, navigation bar, notch or keyboard, or targeting Android 15 and later.

ABAP 开发踩坑与解法库（跨项目通用）。沉淀自示例客户 DEV ERP 项目 100+ 份交付报告与逐日纠错记录，共约 120 条「现象 → 根因 → 解法」，另附新项目开工 checklist（摸底/动手前/编码自检/交付/收尾五阶段）。覆盖 ADT 激活假成功/假锁判定、ABAP 语法硬约束、「返回成功但数据没变」的静默失效清单、事务与 LUW、F4 与弹窗与 ALV、接口 JSON 与网关、DDIC 建对象、ST22 与探针排查手法、打补丁脚本陷阱。开新 ABAP 项目做开场阅读与开工检查，或遇到「编译通过但激活失败 / 激活成功但运行不对 / 返回 S 但值没变 / 报错文案与真实原因无关 / 按钮点了没反应 / 数据莫名错位」时加载本技能。

Help users build interactive graphics applications with LeaferJS (Leafer), from a plain-language idea to a runnable preview. Use when users mention LeaferJS or Leafer AI, or want to create a canvas editor, infinite canvas, whiteboard, design tool, image or poster editor, flowchart, mind map, AI canvas, industrial visualization, interactive animation, or 2D game, even if they have not chosen a graphics engine. Also use for LeaferJS API questions, debugging, performance optimization, and plugin development. Recognize requests such as 无限画布、图形编辑器、海报编辑、流程图、思维导图、工业组态、互动动画 and AI 分镜画布.