Coding & Development

Reduce unnecessary test runs and agent context usage while preserving required verification. Use when planning, running, or improving verification for code changes, or when optimizing test reporting and logs.

Architects Swift server backends as gRPC microservices behind a Hummingbird gateway, with one Swift package per service, a Core of typed-throws use cases, an adapter target per technology (Postgres with row-level security, gRPC, Temporal, Valkey) and an executable composition root. Use when planning a Swift backend or splitting it into services, drawing module boundaries or writing Package.swift, adding a service, endpoint, use case, repository, migration, multi-tenant table or durable Temporal workflow, wiring configuration, graceful shutdown, mTLS or JWT auth, naming server types and targets, setting up a package's CI/CD on GitHub Actions, or reviewing a Swift server's architecture or code style, even when the user doesn't name the architecture.

Verifica se uma mudança de código quebrou algo. Modo completo roda tipos, lint, build e testes e compara com um baseline salvo (separa erro NOVO de dívida antiga, e avisa de rota, dependência ou migration nova, destacando migrations com palavras de segurança). Modo rápido (--quick) roda só tipos mais lint e testes dos arquivos alterados. Use no momento certo (ver "Quando rodar"); é obrigatório antes de commit, push e deploy.

Use when the user asks for an adversarial code review of a branch or codebase, interactive triage of Conventional Comments, or coordinated subagent fixes collected in one pull request. Also use when the user asks to delete or clean up review-work page chats.

Design or challenge software architecture from concrete flows, invariants, ownership and failure evidence. Use for new system design, unclear module boundaries, cross-cutting bug clusters, or changes to state, authority, history or external-effect contracts. Keep ordinary local fixes local; this is not a mandatory architecture ceremony for every edit.

Создание, обновление, исследование, тестирование, ревью, упаковка и улучшение навыков (скиллов) ИИ-агентов в формате Agent Skills и SKILL.md для Codex, Claude Code, Hermes и совместимых сред, включая MCP/API-интеграции и проблемы активации.

Turn your app into a plan that can grow from 5 users to 5,000 users, in plain words. Upload your app idea or your app code, answer a few questions, and get a full scale readiness check, with a database index list, a caching plan, a load test plan, and a hosting plan, for any AI agent. Structured scale plan with database indexing, caching, load testing, and hosting, ready to brief Claude, Cursor, or any AI builder, or a human developer.

Use when the user says "harden my app", "before launch", "someone could drain my API key", "add rate limits to my AI endpoint", "abuse controls", "free tier on my key", "bring your own key" or "spend cap", or is about to make an AI demo public that calls a paid AI API with the owner's key, on Cloudflare Workers, Vercel, Netlify, Cloud Run or similar. Runs one hardening round (fail-closed quotas, keyed IP hashes, a bot gate, CSP, provider terms, spend-cap arithmetic) and returns a findings table with tested patches.

Performs comprehensive, evidence-driven audits of Next.js applications. Evaluates architecture, Next.js best practices, React component design, security, business logic, testing quality, performance, and dependencies. Produces persistent findings, coverage records, and an actionable audit report.

Assess application security before production or releases. Use for requested security audits and release-readiness reviews; deliver a Markdown report with evidence, graphs, and a remediation plan.

Review patches for necessary scope, preserved behavior, and simpler design. Use for code review or pre-delivery self-review.

Senior frontend engineer for web and mobile (React, React Native / Expo, TanStack Query, TypeScript): bundling, lazy loading, micro frontends, accessibility, localization, REST consumption. Builds screens and components from the designer's handoff contract and the architect's API contracts. Use 'frontend setup', 'frontend build <SCR-id | feature | task>' or 'frontend review <target>'.

Use when a jax.profiler.trace() or XProf capture (xplane.pb, trace.json.gz) needs a deep investigation rather than a quick verdict - what to change to make a step faster, the operators on each stream of each GPU, GPU kernels tied back to HLO ops and Python source lines, critical path and stragglers across GPUs, or where the trace can't be trusted - including HIP/CUDA graph (XLA command buffer) kernels labelled command_buffer_N.

Parceiro técnico com a mentalidade de um Analista de Testes Sênior para aplicações web modernas (SPA, APIs, microsserviços, assíncronas) e legadas (JSP, JSF, ASP.NET Web Forms, PHP tradicional, monolitos). Use sempre que o pedido envolver planejar, desenhar, revisar, executar ou relatar testes de software; escolher ou combinar técnicas (particionamento, valor-limite, tabela de decisão, transição de estados) e heurísticas (SFDIPOT/HTSM, FEW HICCUPPS, ALTER FACE, Goldilocks, CRUD, tours, RCRCRC, MCOASTER); fazer testes exploratórios, charters ou sessões; analisar riscos; investigar ou isolar um bug; testar sistemas sem documentação; revisar casos de teste; decidir quando parar de testar — mesmo que o usuário não use a palavra "heurística" nem peça explicitamente uma estratégia.

Use when the user asks to execute an approved implementation plan or defined coding work with subagents, isolated Git worktrees, and fast-forward integration, including requests to use the execute-work skill.

Software architect (DDD, SOLID, hexagonal, microservices, cloud). Use 'architect design' to produce the technical solution, or 'architect impl' plus a task to plan, delegate, review and open a PR.

Design, implement or review access policy for protected resources and operations. Use for new protected entrypoints, roles, ownership or tenant scope, field visibility, sharing links, delegated jobs and permission changes. Authentication alone does not settle access; visual-only edits need no authorization review.

Use when a JAX / XLA program ran out of GPU memory (RESOURCE_EXHAUSTED, "Out of memory while trying to allocate", OOM on the first step or only after N steps) or uses more GPU memory than expected, and a jax.profiler / XProf capture (.xplane.pb) of the run is available.

Use when designing, orchestrating, or executing multi-agent iterative loops, complex multi-stage tasks with feedback cycles, or when implementing autonomous self-improving workflows with persistent memory (.loop-memory) and step tracking (etapas.json).

Create evidence-backed AWS-style architecture diagrams in Mermaid, Graphviz DOT, or both. Use when AWS system context, topology, architecture, or data-flow visuals are requested. Select the format from the required renderer, audience, icon fidelity, layout control, reviewability, and output artifacts; neither Mermaid nor DOT is inherently preferred.

Compaction-immune, DAG-based ticket resolution workflow using Stuart Feldman's Makefile architecture, persistent Memento state files, subagent isolation, 27-language polyglot TDD, design pattern guides, adversarial review, and human pause gates. Use whenever resolving tickets, implementing features, fixing bugs, or executing structured coding workflows.

Turn a 2D house plan into a photoreal 3D house in Blender Cycles, checked by overlaying the build on the drawing. Stills, a film of the walls rising off the drawing, and whole streets of houses. Use for a floor plan, a house plan or a blueprint that should become a real-looking 3D house or street, "từ house plan làm nhà 3D", "bản vẽ thành nhà 3D", "dựng nhà từ mặt bằng", "nhà 3D từ bản vẽ".

为英文产品片、发布会式 demo、黑客松视频制作接近专业解说员水准的 AI 旁白。按句精修、配合画面时长、混音不“泵”，并用可量化的方法判断“听起来怪不怪”。

Review manufacturing-ready mechanical part models, engineering drawings, and assemblies. Covers ISO fits and tolerances, general tolerances (ISO 2768), ISO 20457 plastic molding tolerances, GD&T (ISO 1101 / ASME Y14.5), tolerance stack-ups, and process-specific DFM for injection molding, sheet metal fabrication & cutting, machining, additive manufacturing, and progressive stamping. Use when a user asks to inspect, review, assess, or find issues in a drawing, part, model, or assembly; check tolerances, fits, bearing fits, datums, tolerance stack-ups, or manufacturability.