Skills
96,981 packaged capabilities and playbooks for agents. Install any of them with one command.
Agent workflows
Migrate an existing repository from the legacy planned-features.md plus per-feature design.md/tasks.md workflow to dstack's Copier- and Beads-based workflow. Use when asked to adopt dstack, convert legacy workflow state, or resume an interrupted workflow migration.
(NS) Interview users through focused technical questions to recommend LangGraph vs CrewAI, propose multi-agent architecture, and design agent personas with tools and models. Use whenever the user wants to build a multi-agent system, choose between LangGraph and CrewAI, design agent crews or graphs, plan human-in-the-loop workflows, or asks how to orchestrate AI agents — even if they only say "I want a bot for my company" or "agents working together". Do NOT use for general app requirements (use pm-clarify-requirements) or implementation coding without architecture intent.
Design and implement multi-agent LLM architectures using the orchestrator-subagent pattern. Use when: (1) Deciding whether to use multi-agent vs single-agent systems, (2) Implementing context isolation for high-volume operations, (3) Parallelizing independent research tasks, (4) Creating specialized agents with focused tool sets, (5) Building verification subagents for quality assurance, or (6) Analyzing context-centric decomposition boundaries.
Review code changes through multiple AI models before sending to human reviewers. Use when the user asks to review a branch with sub-agents, run a multi-model code review, get AI review feedback, or mentions "review my branch," "sub-agent review," or "iterate on the code."
Guide to the OpenAI Agents SDK for JavaScript and TypeScript (@openai/agents, @openai/agents- realtime), covering text agents and realtime voice agents. Covers agents, tools defined with Zod schemas, multi-agent workflows and handoffs, input and output guardrails, structured outputs, streaming, human-in-the-loop patterns, and integrations for Cloudflare Workers, Next.js, React, Hono and Node.js, with templates for each agent type plus error handling and debugging strategies. Prevents 9+ documented errors including Zod schema type errors, MCP tracing failures, agent infinite loops, tool call failures, and schema mismatches. Use when building or debugging agent workflows with this SDK, adding tools or handoffs, implementing realtime voice agents, or hitting those errors.
Iteratively optimizes prompts for token efficiency by maximizing information density - reduces verbosity, removes redundancy, tightens phrasing while preserving semantic content. Use when asked to compress, shorten, reduce tokens, tighten, maximize density, increase information density, or make a prompt more concise.
Use this skill whenever the user wants to peer program, pair program, or do collaborative coding where THE HUMAN writes the code and Claude only assists. Trigger on phrases like "pair program", "peer program", "let's code together", "I'll write you review", "drive me through this", "I want to be the driver", "you navigate I type", "teach me by letting me write it", or any workflow where the human is the typist and Claude is the navigator/reviewer. Also trigger when the user wants to learn by writing code themselves with Claude's guidance, or explicitly says Claude should NOT write the implementation. Use this even if the user doesn't say "skill" — the inversion of the usual write/review roles is the signal.
Reviews a GitHub pull request and posts inline comments plus one consolidated summary, adapting to any codebase by discovering the project's own test runner, requirement specs, and architecture conventions before running six specialized review agents in parallel. Stack-agnostic across language and framework; targets GitHub PRs via the gh CLI. Use when the user says "review PR 128", "review this PR", "code review this PR", or "check this pull request". Do NOT use for creating PRs or responding to review comments (use gh-address-comments), or debugging failing CI checks (use gh-fix-ci).
Build Claude prompts via an interview grounded in Anthropic best practices. Use when creating, writing, improving, or fixing a prompt—or when the idea is still vague. Triggers: write a prompt; create a system prompt; my prompt isn't working; help me prompt Claude.
Expert prompt engineering skill that transforms Claude into "Alpha-Prompt" - a master prompt engineer who collaboratively crafts high-quality prompts through flexible dialogue. Activates when user asks to "optimize prompt", "improve system instruction", "enhance AI instruction", or mentions prompt engineering tasks.
Python SDK for inference.sh - run AI apps, build agents, and integrate with all models. Package: inferencesh (pip install inferencesh). Supports sync/async, streaming, file uploads. Build agents with template or ad-hoc patterns, tool builder API, skills, and human approval. Use for: Python integration, AI apps, agent development, RAG pipelines, automation. Triggers: python sdk, inferencesh, pip install, python api, python client, async inference, python agent, tool builder python, programmatic ai, python integration, sdk python
Hardening a skill so it resists rationalization under pressure: run combined-pressure scenarios on a subagent without the skill (RED), capture excuses verbatim, write/edit the skill (GREEN), then plug loopholes (REFACTOR) until compliance holds. Use when deploying or editing a discipline-enforcing skill that an agent could rationalize away. Skip for pure reference skills with no rules to violate.
Get in the jeep and drive across the savanna, stopping at each item in a collection to observe, document, and design improvements. The immersive, adventure-driven way to systematically review and polish a set of related things. Use when auditing, reviewing, or planning polish for a group of pages, components, curios, endpoints, or anything with N items to visit.
Expert integration for the Serena MCP Server - a powerful coding agent toolkit providing IDE-like semantic code understanding to LLMs. This skill should be used when working with codebases through Serena tools, setting up Serena projects, performing semantic code navigation and editing, managing project memories, debugging complex automation workflows, or integrating Serena with Claude Desktop, Claude Code, Codex, ChatGPT, or custom agents. Triggers on Serena tool usage, project activation/onboarding, symbolic code operations (find_symbol, replace_symbol_body, etc.), memory management (write_memory, read_memory), and MCP server configuration. Use for large/complex codebases requiring structural understanding, refactoring tasks, and token-efficient code operations.
Create or reconcile a repository's root AGENTS.md references to the tech-preferences, engineering-principles and verifiable-trust skills. Use when asked to set up, refresh or validate James's personal engineering guidance; preserve existing instructions and make repeat runs idempotent.
Take any user output — prompt, story, procedure, analysis — and make it maximally effective through iterative intent extraction and focused questioning. Use when user says "sharpen this", "make this more effective", "help me clarify this", or "refine my writing".
An integrated skill architect and format review tool that combines the official skill-creator workflow, the former skill-lint compliance checks, and release-spec synchronization. Use this skill when the user needs to create a new skill, edit an existing skill, package a skill, review skill format compliance, check Frontmatter/version/documentation consistency, or generate a skill audit report. Do not use for: creating non-Claude Code skills, code review, functional testing, or general programming tasks.
Create new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, update or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy.
Use when reviewing, grading, critiquing, or improving an existing agent skill or SKILL.md. Checks name/scope, invocation description, structure, leanness, and surfaces concrete opportunities to add references, scripts, and determinism.
Overall controller/director for video creation. Chains a video from "reference material → topic selection → script → editing → B-roll → subtitles → cover" into one pipeline, decides which step to take and which space-video-* sub-skill to call, and reviews the work after the final cut. Triggers when the user says "make a video", "I want to shoot a talking-head piece", "help me do this video from scratch", "video workflow", "video pipeline", or is unsure which video sub-skill to use.
The spec-driven-tdd delivery workflow (orchestrator) - runs the full lifecycle for an OpenSpec change: OpenSpec planning/tracking + Superpowers TDD, simplify, and code review. Invoked by using-spec-driven-tdd; run it once implementation starts
Cross-repo migration swarm — one coordinator + N parallel subagents (one per target repo) that apply the same transformation, open PRs, wait for CI, and report back to a shared JSON ledger. Coordinator handles topology, conflict auto-rebase, and stop-on-novel-failure. Use when bumping a shared dependency, rolling out a workflow change, or applying a codemod across the org. Do NOT use for single-repo work — that's /ork:implement.
Run a local-first task board for people and agents: SQLite cards with a markdown mirror, agent claims with heartbeats, a read-only web board, and an MCP server. Use when the user asks to work with the local task board: cards, claims, or the board itself. Russian discovery terms: доска задач, канбан, задачи для агентов, вести задачи,
Use when tasks need real-browser web automation in Chrome/Chromium via CDP: open or navigate URLs, click/type/select in forms, run page JS, wait for selectors, scrape structured content, capture screenshots, validate UI flows, or run measured web-browser latency checks (`bench:eval`, `bench:all`) for perf regressions.
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.