Skills
97,725 packaged capabilities and playbooks for agents. Install any of them with one command.
Development
Delete the comments in a diff that the code should make unnecessary, through a fresh read-only reviewer, then fix the code each deleted workaround comment was excusing. Use before a change goes to review, or when the user asks to strip comments from a diff.
Runs shell commands in the background with a fire-and-forget approach and delivers progress notify turns every 2 minutes plus one notify turn to the agent session when each finishes, through the `agent-shell` command (a separate Bun project, installed to `~/.local/bin`). Use when a long command, build, test run or server must keep going without blocking the session, with the result read only after the completion notify turn arrives. Do not use for a quick command that finishes inside one turn.
Sketch types, signatures and module structure before writing code, from several candidate designs on different models, then implement against the chosen sketch. Use when the user asks to architect or design something, or when a skill such as figure-it-out calls for it.
Design and change UI through an interactive shell. Turn a spec and a low-fidelity sketch, or an existing component, into a high-fidelity mock.html in the project's own style, iterate on it with the user, then implement it. Alternatively, open the shell over the running app so the user can inspect, comment on and restyle real elements, with changes going straight into code and tests. Use when the user invokes /building-ui, asks to mock up a new component or UI element from a sketch, change a component through a mock, implement an approved mock, or review or change the UI of the running app.
Implement and debug research code, select suitable codebases, reproduce baselines, map mechanisms into model architecture, and optimize training from measurements. Use for 科研编码, 论文复现, codebase选型, 模型实现, 训练调试, 显存/吞吐优化, and research dependency conflicts. Experiment protocols belong to designer; campaign coordination belongs to orchestrator.
核验 C 代码里的 MCU 寄存器写入是否与芯片 SVD 一致：外设、寄存器、位域是否存在；值是否超位宽；枚举是否合法；是否写只读位；是否访问前未开时钟；是否硬编码地址。按“报 / 未判定 / 通过”分桶输出，并可生成复查清单。Use when reviewing embedded C init code, register writes, or AI-generated firmware registers; verify register writes against a chip SVD. 关键词：寄存器核验、寄存器检查、AI 生成代码、驱动复查、初始化代码检查、对拍、register check、verify register writes。不用于：HAL/LL 结构体配置语义（会进未判定）、外挂器件命令（Flash、传感器）、时序正确性、通用代码风格、寄存器事实查询（用 embedded-register-lookup）。
Design and run a bespoke, auditable workflow for a task no narrower skill fits, such as a large migration, an ambitious multi-part change, or work the user reviews after stepping away. Scales rigor to the task, runs a hypothesis loop, and logs decisions. Use when the user asks you to figure it out, or for a large task no other workflow covers.
Delegate implementation work to external coding-agent CLIs (OpenAI Codex `codex exec`, Google Antigravity/Gemini `agy -p`, or any harness with an adapter) instead of Claude subagents, so heavy coding burns other providers' quota while Claude acts as foreman — planning, briefing, reviewing diffs and verifying. Use when the user invokes /foreman, asks to “use codex/gemini/agy as workers,” “save Claude quota/limits,” or “let another model do the coding while you review.” Workers run in isolated git worktrees via scripts/dispatch.sh.
Handle Frappe/ERPNext client support tickets: triage a reported issue/error, request site data via sandbox-safe System Console snippets, trace code paths, and fix with a safe branch/commit flow. Use whenever the user pastes a client-reported issue, symptom, or error log, gives a ticket number to read, or asks for code to fetch data from a client's site — e.g. "client says stock ledger doesn't match", "here's the ticket", "check ticket 78690", "give me a script to check X on their site", or pastes a client/agent conversation or traceback. Also use to refresh context from the latest communication — "check the latest client reply", "any new reply on the ticket", "refresh the thread"
Make a bug in a TypeScript resource executable before fixing it, with a regression test that fails before the fix and passes after. Use only when the user asks for TDD, a failing test, or a regression test, or when the resource already has a test runner and the bug has an obvious cheap test target. Skip it when the test path is unclear, expensive or integration-heavy.
TypeScript best practices for FiveM resources and their NUI. Use when reading or editing any .ts or .tsx file or a .svelte file with <script lang="ts">, or when the user asks how to type net events, callbacks, exports, NUI messages or state bags.
Find why code is the way it is, through read-only investigators that search git history, PRs, issues and any docs or chat the agent can reach, then return a cited, confidence-weighted answer. Use when the user asks why something works this way, why a choice was made, where a number came from, or what a regression or postmortem traces back to, or when a skill such as architect calls for it.
Otimiza qualquer landing page ou aplicação web (React, Vite, Next.js, HTML) para atingir nota máxima (98-100) no Google Lighthouse 13 em Performance, Accessibility, Best Practices, SEO e Agentic Browsing, aplicando técnicas avançadas de SEO, AEO (Answer Engine Optimization) e GEO (Generative Engine Optimization para ChatGPT, Perplexity, Claude e Gemini). Use quando o usuário pedir para melhorar SEO, AEO, GEO, ser encontrado por IAs/buscadores, criar llms.txt/ard.json ou bater nota 100 no Google Lighthouse.
Use GitLab from the terminal without registering an MCP server. Covers authenticating the zereight-mcp-gitlab CLI (PAT or OAuth device flow) and reading merge requests, diffs, discussions, issues, and pipelines. Use when the user asks about a GitLab MR/issue/pipeline and no GitLab MCP tools are available, or wants CLI-only setup.
SSoT design draft: clarify the objects, permitted changes and invariants, and separate architectural and business responsibilities before writing the formal VDM. Read ssot first, then this skill; use for drafting or restructuring a design, not for implementing, checking or reviewing specifications.
Build two-way Google Calendar sync for a booking or scheduling app: hosts connect their Google account with OAuth, events on the calendars they choose block their availability, kept current by push notifications and a renewal cron, and confirmed bookings are written to their calendar as events, with Google Meet links when asked. Use when: (1) an app with bookings, appointments, classes or shifts must respect staff's Google Calendar busy time or put bookings into it, (2) an existing calendar sync shows phantom busy time, missed or duplicate events, group sessions blocked after the first booking, sync that stopped without an error, or users stuck after revoking access, (3) the user mentions: Google Calendar integration, two-way calendar sync, connect Google Calendar, busy time, free/busy, events.watch, push notifications, X-Goog-Channel-Token, channel renewal, calendar.events scope, refresh token, invalid_grant, offline access, PKCE, Google Meet link, conferenceData, extendedProperties, hangoutLink. Carries a dependency-free Google client on fetch, a pure busy-time mapper and diff, single-flight sync, idempotent booking events, make-before-break channel renewal, derived health per calendar, Firestore and Supabase stores, and a 32-test node:test suite run under four process time zones. Next.js App Router route handlers, with an Express adapter for other Node servers; storage, auth and the booking source are seams. Not Outlook or Microsoft 365 (microsoft-calendar-sync), not slot generation (time-slot-generator), not a calendar UI, not ICS feeds
Build two-way Microsoft 365 / Outlook calendar sync for a booking or scheduling backend over Microsoft Graph: hosts connect their Microsoft account with OAuth, events on chosen calendars block their availability (change-notification subscriptions plus a renewal cron), and confirmed bookings are written to their calendar as events, with Teams meeting links. Use when: (1) an app with bookings, appointments, classes or shifts must respect staff's Outlook busy time or put bookings into their Outlook calendar, (2) an existing Outlook or Graph calendar integration stops working an hour after connecting, never receives notifications, ignores vacations, duplicates events or loses its subscriptions, (3) the user mentions: Outlook calendar integration, Microsoft 365 calendar, Office 365 calendar sync, Microsoft Graph calendar, Entra ID app registration, Calendars.ReadWrite, calendarView, change notifications, Graph subscription, clientState, validationToken, lifecycleNotificationUrl, subscriptionRemoved, MSAL refresh token, AADSTS700082, invalid_grant, interaction_required, singleValueExtendedProperties, teamsForBusiness, showAs, isAllDay, mailboxSettings timeZone. Carries a dependency-free Graph client on fetch, a pure busy-time mapper that anchors all-day events to the mailbox zone, single-flight sync, booking events found again by an extended-property marker, in-place subscription renewal that replaces lost subscriptions, derived health per calendar, Firestore and Supabase stores, and a 43-test node:test suite run under four process time zones. TypeScript for Next.js App Router route handlers, with an Express adapter for any other Node server; storage, auth and the booking source are seams behind CalendarStore, authenticate and BookingSource. Not Google Calendar, not Exchange on-premises (EWS), not a calendar UI.
This skill should be used when the user asks to "create MongoDB model", "define Beanie document", "write MongoDB query", "create aggregation pipeline", "run database migration", "index MongoDB collection", or mentions Beanie, Motor, MongoDB documents, or async database operations. Provides MongoDB/Beanie ODM patterns for FastAPI.
This skill should be used when the user asks to "implement FSD", "use Feature-Sliced Design", "organize architecture", "structure project folders", "set up FSD layers", "create feature slices", "refactor to FSD", "add FSD structure", or mentions "feature slices", "layered architecture", "FSD methodology", "architectural organization", "views layer", "entities layer", "shared layer", "Next.js with FSD", or "Turborepo FSD structure". Provides comprehensive guidance for implementing Feature-Sliced Design methodology in Next.js applications with custom 'views' layer naming.
Comprehensive UX design patterns for Rails applications including responsive design, animations/transitions, dark mode, loading states, form UX, and performance optimization. Use this skill when implementing user-facing features requiring polished interactions and responsive layouts. Trigger keywords: UX, responsive, mobile-first, animation, transition, dark mode, loading, skeleton, progress, toast, form validation, performance, Core Web Vitals, user flow
New to MCP?
From what the protocol is to installing your first server, these answers cover most first questions.
An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.