agentic-fullstack-workbench

Single entry point for .NET/C# backend microservice work — C# design & async, HotChocolate GraphQL (ChilliCream), MassTransit + Azure Service Bus messaging, MongoDB data access, service startup/DI/Confix, OpenTelemetry, scaffolding a new service, and strict backend code review. Use when: implement backend feature, GraphQL resolver/ObjectType/TypeExtension/DataLoader/mutation convention/stitching/Fusion gateway/secure IDs, IConsumer/publisher/request-response, MongoDB repository, startup/DI/health checks, SOLID/async/await/CancellationToken/records/ValueTask/error handling/performance, scaffold new microservice (Abstractions/Core/DataAccess/GraphQL/Host/Worker), or backend/PR/architect code review.

Single entry point for building CopilotKit v2 AI features — add/bootstrap CopilotKit, React UI (@copilotkit/react-core/v2: provider, CopilotChat/Popup/Sidebar, useAgent, useFrontendTool, useRenderTool, human-in-the-loop, threads, renderers), server runtime (@copilotkit/runtime/v2: CopilotRuntime, createCopilotRuntimeHandler, BuiltInAgent, AgentRunner, middleware, server tools, transcription), the AG-UI protocol (SSE, AbstractAgent/HttpAgent, state sync, tool calls), external agent-framework integrations (LangGraph, CrewAI, PydanticAI, Mastra, ADK, LlamaIndex, Agno, Strands, Microsoft Agent Framework), A2UI declarative surface rendering, and debugging. Use when: install/wire CopilotKit, build a chat/copilot feature, register frontend/server tools, share app context, handle agent interrupts/HITL, implement or wire an AG-UI agent backend, render A2UI surfaces, or debug runtime connectivity/streaming/tool/transcription/version issues. NOT for React UI implementation itself (use frontend-developer) or backend micr

Single entry point for Data & Analytics Platform work — MongoDB/SQL live analysis and query/index optimization, data-pipeline code, Databricks/DAP Lakehouse, and PowerBI. Use when: inspect MongoDB or SQL data read-only, analyze an execution plan/index/query, implement a change tracker/data loader/stored procedure/repository/Squadron DB test, build a Bronze/Silver/Gold Delta pipeline or PySpark notebook, design a Lakehouse layer, or create a PowerBI model, DAX measure, Power Query, or dashboard. NOT for backend microservices (use backend-developer).

Single entry point for frontend web/mobile development — React 19 + TypeScript + Vite engineering, Relay GraphQL client (fragments/queries/mutations + performance), React performance, composition patterns, React Native/Expo, View Transitions, Formik/Yup forms, Component Design System + CSS Modules UI, UX/accessibility/calm design, web-design audit, mandatory docs research, and frontend validation/quality. Use when: implement/refactor/debug a React or React Native feature, TypeScript UI, Vite config/plugin/SSR, Relay fragment/query/mutation/pagination/fetch-policy, re-render/bundle/waterfall performance, compound components/render props/component API, View Transition animation, Formik form, CSS/layout/spacing/color/visual polish, UX flow/accessibility/AI-UX, a11y/design audit, or validating frontend changes (typecheck/Biome/Vitest/Relay compiler/build/browser). NOT for CopilotKit (use the copilotkit-developer skill) or E2E/browser test automation (use test-automation-engineer).

Single entry point for acceptance testing and UI test automation — derive deterministic manual cases and coverage from OpenSpec scenarios, prepare explicitly approved ADO Test Plan cases, and implement Playwright E2E/visual tests with strict Page Object Model, BrowserStack, CLI/MCP browser automation, and local webapp validation. Use when: assess requirement testability; design ISTQB-style positive, negative, boundary, decision-table, or state-transition cases; create a coverage matrix or ADO test case; create/fix a Playwright spec or POM; diagnose flakes/selectors; run visual, cross-browser, BrowserStack, console, network, or local browser checks. NOT for React/UI implementation (use frontend-developer) or backend tests (use backend-developer).

Single entry point for authoring and curating Copilot context — create/update skills, agents, instructions, prompts, and project memory; consolidate overlapping customizations; debug discovery; and build compact DIGEST/RAW knowledge. Use when: create or review SKILL.md, .agent.md, .instructions.md, .prompt.md, AGENTS.md, or .github/memory; design routing, handoffs, applyTo patterns, tool restrictions, progressive disclosure, prompt structure, context curation, or Hierarchical RAG.

Ultra-compressed communication mode. Cuts token usage ~75% by speaking like caveman while keeping full technical accuracy. Supports intensity levels: lite, full (default), ultra, wenyan-lite, wenyan-full, wenyan-ultra. Use when user says "caveman mode", "talk like caveman", "use caveman", "less tokens", "be brief", or invokes /caveman. Also auto-triggers when token efficiency is requested.

Safe, concise Git workflow guardian for merges, rebases, cherry-picks, branches, commits, conflicts, recovery, release tags, and repository hygiene. Loads detailed references only when needed.