Skills
96,770 packaged capabilities and playbooks for agents. Install any of them with one command.
Testing & quality
Integrate the ONE store H5 Game SDK (Web SDK) into an HTML5 web game, check the integration status, and review the code before release. Use when the conversation mentions ONE store, OneStore, 원스토어, web game, 웹게임, HTML5 game, H5, WebView game, in-app purchase, 인앱결제, ads, 광고, rewarded or interstitial ads, createSDK, ONEstoreH5, initializeAsync, startGameAsync, purchase, getProductDetailsAsync, showRewardedAsync, or an iframe-hosted game, or when the user sends a command word such as doctor, integrate, iap, ads, embed, fix, or check to this skill. Covers loading the SDK, the boot flow (init, loading, start), lifecycle handling, ad and purchase code, iframe embedding requirements, error (reason) diagnosis, a blank screen in the app, and pre-release checks. When asked to write ad, purchase, or lifecycle code for a ONE store web game, use this skill instead of writing it from scratch, because it contains reference code based on the official sample.
Generate safe, consistent mock sample data that matches the current project's schema and business rules. Use when the user needs to check screens, flows or tests without real data, with requests such as "make dummy data", "fill in test data" or "create demo cases". Do not use to transform real personal information or company secrets, or to produce operational data.
Verify with real behavior and evidence whether the current project meets the user's request and confirmed completion criteria. Use for requests such as “Does it work properly?”, “Is it really done?”, “Test it,” “Is it ready to deploy?”, or “Recheck the changes.” Do not use for requests to change functionality or design during verification.
Runs diagnostics and a health check on one PostHog experiment: its setup, exposures, results and changes during the run. Covers 0 exposures, sample ratio mismatch, lost or uneven exposures, users in multiple variants, identity faults, who the experiment counts, significance traps (peeking, A/A, 'target reached', Bayesian vs Frequentist, sequential, CUPED), mid-run edits, pause and freeze, and survey follow-up. TRIGGER when: user asks 'is my experiment healthy / set up right / biased?' or 'why 0 exposures?', asks why it counts different people than their insight, mentions the bias or mismatch warning, an uneven split, significance that flips, numbers that disagree with their SQL, results that changed after an edit, or wants user feedback on an experiment. DO NOT TRIGGER when: creating an experiment (use creating-experiments), only configuring rollout (use configuring-experiment-rollout) or metrics (use configuring-experiment-analytics), or only asking lifecycle questions (use managing-experiment-lifecycle).
Inspect and compare offline AI evaluation experiments, diagnose case-level regressions, and follow scorer history across application, model, or prompt changes. Use when a user asks whether an offline run improved, why a case failed, how scores changed between runs, or how to publish externally computed offline results. Covers immutable scorer versions, comparable dataset cohorts, incomplete coverage, and selective payload reads.
Prepare, verify and attest a first-party Skills release, or audit an existing tag and its installation evidence. Use for a new release, fresh-install verification or release-evidence closeout; execute only the requested stages and authorized targets
Audits whether AGENTS.md and/or CLAUDE.md in the current repository still reflect the project's reality (structure, commands, conventions, scope) and reports discrepancies without editing them. Use when the user runs /audit-agent-context, asks to verify or update agent context documentation, or suspects AGENTS.md/CLAUDE.md is outdated
Whole-codebase deep audit. Partitions the repo into 25-file components, runs 9 perspectives by 3 models (OpenAI, Gemini, Claude) on every component via opencode run with Serena navigation, sweeps for gaps, and runs one ocr scan coverage pass. Then collates and deduplicates the findings and verifies every Critical and High issue against a clean export of the reviewed commit with isolated claude -p judges. The run directory feeds the review-tickets skill.
Turn verified issues from a review-deep, review-full, or review-diff run into Jira ticket files. Re-cuts each issue into one ticket per file from the reviewers' own text, judges and edits every finding against a clean checkout with isolated claude -p processes, merges tickets that share one fix, assembles Jira wiki tickets with code copied from git, and checks them. Ends with posting guidance for any Jira client and a bundle ticket for deferred low-priority work.
Review a web interface, a UI diff or a single component and return ranked, fixable findings. Use for design critique, UI audits, accessibility checks, reviewing a branch or PR for interface impact, stress-testing one component across states and sizes, or explaining how an observed interface or animation is built
Review the changes since a fixed point (commit, branch, tag, or merge-base) along two axes: Standards (does the code follow this repo's documented coding standards?) and Spec (does the code match what the originating issue/spec asked for?). Runs both reviews in parallel sub-agents and reports them side by side. Use when the user wants to review a branch, a PR, work-in-progress changes, or asks to "review since X".
Test how coding agents discover and use your product. Use when asked to run agent evaluations, test whether agents can use an SDK, API, CLI, or MCP server, find where agents struggle with onboarding, or measure whether docs and skills improve agent success. Runs real agent sessions and scores outcomes through the Gauge CLI. Also use for Gauge Agents tasks; not ordinary unit tests or AI-search visibility tracking.
Draft 2–3 firegoat evals (firegoat.yaml) for the current repo: tests that a coding agent runs locally with `npx -y firegoat@latest eval`, each with a prompt, an end_goal, and script checks. Use when the user says "write firegoat evals", "draft evals", "set up firegoat", "firegoat init", or asks for local agent-experience (AX) evals of their SDK, CLI, API, or docs.
Game QA testing with Playwright — visual regression, gameplay verification, performance, and accessibility for browser games. Use when writing or running game tests, debugging test failures, or building QA infrastructure. This is the reference skill — use qa-game for the user-facing command.
Analyze a game, find what needs work, and implement the highest-impact improvements. Use when the user says “improve my game,” “make my game better,” “fix my game,” “what's wrong with my game,” or “polish my game.” Run repeatedly—each pass finds the next most impactful thing to fix. Do NOT use for adding specific new features (use add-feature) or initial game creation (use viral-game for one-shot builds, make-game for milestone-driven projects)
Review an existing game codebase for architecture, performance, and best practices. Use when the user says "review my game", "code review", "check my game architecture", "is my game well structured", or "audit my game code". Do NOT use for making changes — this is read-only analysis. Use improve-game to implement fixes.
Tests in real browsers via Chrome DevTools MCP. Use when building or debugging anything that runs in a browser. Use when you need to inspect the DOM, capture console errors, analyze network requests, profile performance, or verify visual output with real runtime data. Requires the chrome-devtools MCP server to be configured.
Conducts multi-axis code review. Use before merging any change. Use when reviewing code written by yourself, another agent, or a human. Use when you need to assess code quality across multiple dimensions before it enters the main branch. Use when asked to review a diff or a pull request, even when the diff is pasted inline.
Review the changes since a fixed point (commit, branch, tag, or merge-base) along two axes: Standards (does the code follow this repo's documented coding standards?) and Spec (does the code match what the originating issue/spec asked for?). Runs both reviews in parallel sub-agents and reports them side by side. Use when the user wants to review a branch, a PR, work-in-progress changes, or asks to "review since X".
Subjects every non-trivial decision to a fresh-context adversarial review before it stands. Use when you want every assumption cross-examined before proceeding, when stress-testing a plan for hidden failure modes, when correctness matters more than speed, when working in unfamiliar code, when stakes are high (production auth, security-sensitive logic, a high-stakes migration, irreversible operations), or any time a confident output would be cheaper to verify now than to debug later.
Find the mistakes agents keep repeating in this repo and make each one impossible. Try architecture first, then types, then a lint whose error names the fix, then a test, and write docs last. Prove each check fails on a real past mistake. Repeat this each time the operator corrects you. Use for /correct.
Transform mocked-up UI specs (specs-ui skill project) into a purely UI clickable prototype in HTML + Tailwind + Alpine.js: main flows are wired up (navigation, forms, states, simulated data stored in the browser, reset), with guided test scenarios for user testing and demos, published as a multi-page artifact. Use whenever someone wants a prototype, clickable mockup, interactive demo, user-testable flow, user test, or a version "you can click through" based on existing specs or mockups, even without saying "prototype".
Verify web UI changes in headless Chrome with agent-browser, signed in as the project's Clerk test users, and run the project's relevant existing tests. Use after changing anything visible, or when asked to check, inspect, screenshot or exercise a browser flow.
Investigate a LangSmith experiment with the LangSmith CLI, identify examples with poor evaluator results, explain likely causes using run evidence, and propose targeted fixes. Use when a user asks why an experiment performed badly or which evaluated examples need attention.
Find where an agent pipeline can swap frontier models for cheaper ones without losing quality. Drive the installed rightmodeler TypeScript CLI through its resumable pipeline, interpret its machine protocol, present per-family verdicts, and open a reviewed model-swap pull request when the evidence supports one. Use when the user wants to cut model spend, right-size models per task, benchmark cheaper model substitutions, or analyze an agentic pipeline for cost optimization.
Audit the skill inventory for quality, coverage, and staleness. Detect overlapping, incomplete, or outdated skills and produce a health report. Activate when: skill audit, stocktake, skill inventory, check skills, which skills exist, audit skills, skill quality, skill coverage.
Review the diff since a fixed point along Standards and Spec axes in parallel sub-agents, reporting findings without editing the target or deciding the final verdict. Use when the user wants to review a branch, PR, or work-in-progress diff, or when review-loop or project-review requests a specialist software check
Send an independent judge that is not the author: a fresh evaluator subagent that checks finished work against pass criteria written before it was built (evaluator mode), or audits a batch against its source of truth (audit mode), with evidence for every verdict and never a fix. Use when the owner asks for an evaluator, reviewer, auditor or judge, or says 'send a judge', 'evaluate this', 'check my work with a fresh agent', 'get a second agent to review this', 'audit this folder', 'audit my registry', 'verify this against the source of truth', or 'LLM-as-judge'. Runs only when asked; it does not fire on its own at the end of a task. Not for checks a machine can run (tests, schema validation, a search that must return zero), not for hundreds of identical units, and not for fixing anything.
Writes, runs and debugs EventCatalog architecture fitness functions — TypeScript rules in `*.fitness.ts` files that check the documented architecture (ownership, naming, domain boundaries, dependencies, data access, schemas) and fail CI when it drifts. Turns an architecture rule described in plain language into a working fitness function, places it globally or beside the domain, system, service, message or data store it governs, runs `eventcatalog fitness check`, verifies the result, and wires it into CI. Use when user asks to "write a fitness function", "add an architecture rule", "enforce that every event has an owner", "check domain boundaries", "stop services sharing databases", "detect circular dependencies", "fail the build when the architecture breaks a rule", "run fitness checks in CI", or "why is my fitness function failing".
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.
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）。
Adversarial multi-model review of a change. Reviewers on different models challenge it independently, then a lead judgment sorts their findings into act on, consider, noted and dismissed. Use when the user asks to interrogate, stress test, adversarially review or tear apart a change, or when a skill such as architect calls for it.
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.