Skills
96,770 packaged capabilities and playbooks for agents. Install any of them with one command.
Agent workflows
Guide for designing and implementing command-line interfaces (CLIs) that are equally usable by human developers and automated coding agents. Use when the user wants to build a CLI, apply CLI best practices, or use Go with Cobra and Viper.
Orchestrate complex work via a phase-gated multi-agent loop (audit → design → implement → review → validate → deliver). Use when you need to split work into subsystems, run independent audits, reconcile findings into a confirmed issue list, delegate fixes in clusters, enforce PASS/FAIL review gates, and drive an end-to-end validated delivery. Do not use for small, single-file tasks.
Detects whether AI-assisted review tools (such as Luminance, Kira, etc.) are in use, hands off large-batch clause extraction to the tool, and QA's its output according to the trust levels in `~/.claude/plugins/config/claude-for-legal-zh/corporate-legal/CLAUDE.md`. Use when the user says "hand off to an AI tool", "batch review", "AI extraction", or when diligence-issue-extraction encounters large-batch categories.
Multi-perspective code analysis using three AI personas (RYAN, FLASH, SOCRATES) for comprehensive decision-making. Use when complex code decisions need analysis from multiple viewpoints, or when avoiding single-perspective blind spots is critical.
Use when starting a new feature, designing a new project, or making a substantial architectural change to an existing codebase. Entry ramp and spine phase 1. Captures intake before scoped context discovery, then challenges deviations from idiomatic patterns, surfaces refactors as first-class decisions, and produces a validity-checked architecture.md. For epics, decomposes into child changes instead of a single architecture.md. Do NOT use for bugs or tiny changes — use triage instead.
One unified skill for all engineering work: code, UI, docs, and repo maintenance. Auto-detects which specialization to activate from the prompt — ENGINEER for feature work and code changes (full phase-gated SDLC, spec, DAG execution, TDD, evaluator; SPEC-lite mode for design-doc-only handoffs), UI for design and styling (typography, spacing, color, components, dark mode, accessibility), DOCS for documentation (Diátaxis-driven: tutorials, how-tos, reference, explanation, READMEs), MAINTAIN for repo health (security audit, lint/type debt, flaky tests, dead code, dependency upgrades). Blends multiple specializations in a single session when the prompt spans them. Loads only the relevant reference files — a context-efficient alternative to running separate commands. Triggers on any code/feature/build request, "design a UI", "style this", "write docs", "document this", "add a README", "security audit", "fix flaky tests", "upgrade dependencies", or any combination.
Use Beads (bd tool) for dependency-aware task tracking and long-horizon planning in coding projects. This skill should be used when working on complex multi-step projects that span multiple agent sessions, when discovering new work during implementation, or when explicit task dependency management would improve workflow organization and prevent context loss between sessions.
Collaborative requirements discovery session optimized for AI coding workflows. Creates task directories, seeds PRDs, runs codebase research, proposes concrete implementation approaches with trade-offs, and converges on MVP scope through structured Q&A. Use when requirements are unclear, multiple implementation paths exist, trade-offs need evaluation, or a complex feature needs scoping before development.
Use when brainstorming a feature or change whose written design should live in a GitHub issue instead of a docs/ file — including starting, pausing, and resuming a brainstorm across sittings or harnesses, and revising an issue already handed over. Triggers include "brainstorm this into an issue", "spec it out as a GitHub issue", "file an issue for this feature", "continue/resume brainstorming issue #N", "pick up the draft brainstorm", "brainstorming-to-issue #N".
Browser automation via Puppeteer CLI scripts (JSON output). Capabilities: screenshots, PDF generation, web scraping, form automation, network monitoring, performance profiling, JavaScript debugging, headless browsing. Actions: screenshot, scrape, automate, test, profile, monitor, debug browser. Keywords: Puppeteer, headless Chrome, screenshot, PDF, web scraping, form fill, click, navigate, network traffic, performance audit, Lighthouse, console logs, DOM manipulation, element selector, wait, scroll, automation script. Use when: taking screenshots, generating PDFs from web, scraping websites, automating form submissions, monitoring network requests, profiling page performance, debugging JavaScript, testing web UIs.
Operate Circuit, the flow engine that runs coding work as structured, evidence-backed flows (Fix, Build, Explore, Review, Prototype). Use ONLY when the user explicitly invokes Circuit by name — e.g. says "circuit", "run a flow", "use circuit", "start a circuit run", "resume the run", or names Circuit-specific concepts (flows, checkpoints, run folders, the power dial or process override, connectors, `circuit preview`, `circuit reclaim`). Do NOT trigger on generic coding requests, even substantive ones; do NOT trigger merely because Circuit is installed in the project. Absent an explicit mention, do the work by hand.
Delegate a coding or implementation task to a nested Codex CLI agent through unattended, workspace-scoped `codex exec`. For non-Codex harnesses (Claude Code, OpenCode) only — never invoke from Codex itself. Use only when the user explicitly asks to use or delegate to a Codex subagent — not merely because a task involves code.
Multi-agent system research agent for CoWork OS. Use when: researching multi-agent papers, frameworks, production case studies; maintaining a research queue; producing CoWork OS applicability analysis and implementation recommendations. Triggers: 'multi-agent research', 'research multi-agent systems', 'multi-agent papers', 'agent orchestration research', 'CoWork OS research'.
This skill should be used when the user asks to “create an agent”, “make an agent”, “write an agent”, “build a subagent”, “add an agent to a plugin”, “design an autonomous agent”, “generate an agent file”, “write a system prompt for an agent”, “what frontmatter does an agent need”, or “create a specialized agent”. Not for skills or commands—use create-skill.
Create a structured feature specification with self-contained task files organized into parallel execution waves. Use this skill when the user says "create a spec", "plan this feature", "write up an implementation plan", "break this into tasks", or after any planning conversation where the user wants to capture decisions as actionable spec files. Also use when the user says "/create-spec" or wants to decompose a feature into work items that agents can implement independently. This skill produces local spec files under specs/{feature}/ — no GitHub integration.
Before starting a nontrivial task, compute how much understanding the work demands and split it into "already understood" vs "learn first". Use when the user invokes /depth-check, asks "what do I need to understand before touching this?", "이거 시작하기 전에 뭘 알아야 해?", "어디까지 파악하고 들어가야 해?", or is about to start work in territory they may not know well. Maps the task's blast radius, grades the depth each area requires, and emits the short list of questions to answer before coding.
Implementation orchestrator for guardrails projects — dispatch every plan task to a subagent under develop-tdd, merge the task branches, then run one review-conformance review per guidelines file and dispatch the fixes. Use when implementing any feature, fix, or refactor.
Executes exactly one planned Jira workflow task after critique approval. Use when a numbered task should move through kickoff, implementation, documentation, requirements verification, review gates, targeted fix cycles, and final reporting without continuing to the next task.
Build a durable, navigable atlas of a code repository — structure, functional blocks, tech stack, history, CI/CD, strengths and weaknesses, and a "continue development" playbook — plus a distilled summary that bootstraps a future agent. Use when the user wants to understand, onboard to, document, take over, or keep building an unfamiliar codebase, asks for a repo overview / architecture map / codebase walkthrough, or wants to find good issues and PRs to contribute to a public repo.
Advise, prompt, and design workflows for Claude Fable 5. Use when choosing whether Fable is worth its cost, migrating prompts or skills to Fable, setting effort and budgets, routing work across Fable/Opus/Sonnet/Codex, designing long-running agent loops, grounding progress claims, or controlling context/cost for high-autonomy reasoning, coding, research, strategy, and review work.
Delegate a sub-task to Claude Code via the Agent Client Protocol (ACP). Use this skill whenever you want to hand off work to Claude — complex agentic coding with MCP tool access, detailed multi-file refactors, tasks requiring Claude's reasoning style, or anything where Claude's strengths give an advantage. Also invoke when the user asks you to "ask Claude", "use Claude for this", or "run this through Claude". The script handles subprocess lifecycle and ACP session setup; you just provide the prompt and read stdout.
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.