Skills
96,981 packaged capabilities and playbooks for agents. Install any of them with one command.
Agent workflows
Create or update a complete repository skill from a user's idea, including the workflow instructions, references, scripts or assets, agent metadata, skill-card artwork, cinematic banner artwork, README links, discovery metadata, and validation. Use when the user asks to create a new skill, add a skill to this collection, turn a workflow into a reusable skill, or make a skill's documentation and artwork consistent with the repository.
Set up a "software factory" that turns labelled GitHub issues into reviewed pull requests, using coding agents (Claude Code, Codex or any other CLI agent) that run in Upstash Box sandboxes and are started by GitHub Actions. Use this whenever someone wants to build, recreate, extend or debug an issue-to-PR pipeline, a pool of background coding agents, "agents that work on my GitHub issues", a label-triggered agent workflow, or anything that combines Upstash Box with coding agents and GitHub, even if they never say "software factory". Also use it to add a repo, an agent type, workers, a model or effort setting, or a worker image (snapshot) to a factory built this way.
Generate a project-local verification skill that drives your app the way a user does — any language, framework, or platform. Use for /create-verification-skill, "make a control skill for this repo", or when a project has no scripted way to prove UI/CLI/service behavior.
Guide users building apps, scripts, CI pipelines, or automations on top of the Cursor TypeScript SDK (`@cursor/sdk`). Use this skill whenever the user mentions integrating, installing, or writing code against the Cursor SDK; whenever they say `Agent.create`, `Agent.prompt`, `Agent.resume`, `agent.send`, `run.stream`, `CursorAgentError`, or `@cursor/sdk`; whenever they ask to run Cursor agents programmatically from a script, CI/CD pipeline, GitHub Action, backend service, or any other code that isn't the Cursor IDE itself; and whenever they want to pick between local and cloud runtime, configure MCP servers for an SDK agent, or handle streaming, cancellation, or errors from an SDK agent. Also trigger when a user is wiring Cursor into an automation, writing a bot that runs Cursor, or porting REST `/v1/agents` calls to the SDK, even if they don't explicitly name the package. Use this eagerly rather than answering from memory; the SDK surface evolves and this skill plus its references are the source of truth for the external package
Design an agentic AI architecture from a use case and produce a dated design document. Use when the user wants to design or architect an agent or multi-agent system, decide single vs multi-agent, choose where to place memory, human-in-the-loop, or evaluation, or asks how to build an agent for a specific use case.
Use when facing 2+ independent tasks without a written plan, with no conflicting shared mutable state or sequential dependencies, where parallel delegation beats inline cost; otherwise inline. Planned tasks use subagent-driven-development.
Memory consolidation skill for Claude Code. Scans session transcripts for corrections, decisions, preferences, and patterns, then merges findings into persistent memory files. Auto-triggers via native Stop hook every 24hrs. Inspired by how sleep consolidates human memory.
Create and maintain a simple task progress memo in docs/tasks/task-description.md by default, or in an explicit task directory supplied by the user or another active skill. Use when the user wants a plan, TODO steps, acceptance criteria, checkbox syncing, status updates, or progress tracking that stays in sync while the agent works. Treat git commits as required recovery checkpoints: each meaningful change must be committed and the commit id must be written into the matching step note.
Persistent Claude-Codex pairing inside one Herdr tab. Use for live peer work, workflows requesting a pair, or any inbound `[agent ...]` / `[herdr-pair control ...]` line — including after context compaction; resume the exact session and reply through the bundled transport.
Control interactive terminal applications like vim, git rebase -i, git add -i, git add -p, apt, rclone config, sudo, w3m, and TUI apps. Can also supervise another CLI LLM (cursor-agent, codex, etc.) - approve or reject its actions by pressing y/n at confirmation prompts. Use when you need to interact with applications that require keyboard input, show prompts, menus, or have full-screen interfaces. Also use when commands fail or hang with errors like "Input is not a terminal" or "Output is not a terminal". Better than application specific hacks such as GIT_SEQUENCE_EDITOR or bypassing interactivity through file use.
Conventions for the Jev judgment tools. Use when screening fetched or pasted content, verifying claims against evidence, ranking or classifying items by meaning, comparing passages, extracting fields, auditing extracted values, reviewing a patch, gating completion, or judging how likely a proposition is — and when choosing between Jev and rg, curl, or plain reading.
Execute prompts as the Kimi seat through the Pi CLI on OpenRouter, with Moonshot's flagship Kimi K3 model pinned per invocation (moonshotai/kimi-k3) — strong long-horizon coding, large-codebase understanding, 1M-token context, always-on thinking. Use when users explicitly request Kimi execution, when a workflow needs a Kimi-backed seat, or when a cross-runner workflow wants a Moonshot Kimi perspective without leaving the current workspace.
Framework for building LLM-powered applications with agents, chains, and RAG. Supports multiple providers (OpenAI, Anthropic, Google), 500+ integrations, ReAct agents, tool calling, memory management, and vector store retrieval. Use for building chatbots, question-answering systems, autonomous agents, or RAG applications. Best for rapid prototyping and production deployments.
Reviews, curates, and maintains the Forge library of agents, skills, and templates. Performs deduplication analysis, staleness detection, quality promotion, and orphan reference checking. Produces structured review reports with actionable recommendations for merging, archiving, or promoting library items. Use this skill when the user wants to review the library, clean up agents or skills, check what's available, find duplicates, trim unused items, see library statistics, or says "what's in my library?" Also triggers on scheduled review intervals or when the library grows beyond 20 items. Do NOT use for creating new agents (use Agent Creator), creating skills (use Skill Creator), or planning teams (use Mission Planner).
Use when the user asks to build a project end-to-end, run autonomous multi-hour or multi-day development, or orchestrate multi-milestone work without human intervention. Trigger phrases include "build this whole project", "implement this end-to-end", "do this autonomously", "run a long task", "long-running agent", "build it from scratch and don't ask", "/long-task", or "$long-task". Also use for lifecycle subcommands such as `/long-task status`, `/long-task pause`, `/long-task resume`, `/long-task clear`, and `/long-task complete`.
Use when the user asks to "remember project context", review saved findings, initialize runtime memory, archive stale work, reconcile notes, or erase a subject; manages authorized HOT/WARM/COLD working memory across all disciplines while preserving registry event ownership and privacy controls. Not for changing canonical registry facts - route those through the owning registry. 项目记忆/跨会话
Automatically activates when over-engineering patterns are detected: (1) Modifying code or files the user did not explicitly ask to change (2) Creating new abstraction layers (class, interface, factory, wrapper) without being asked (3) Adding comments, documentation, JSDoc, or type annotations without being asked (4) Introducing new dependencies without being asked (5) Rewriting entire files instead of making minimal edits (6) Diff scope significantly exceeding the user's request (7) User signals like "too much", "don't change that", "only change X", "keep it simple", "stop" (8) Adding error handling, validation, or defensive code for scenarios that cannot occur (9) Generating tests, configuration scaffolding, or documentation without being asked
Synthesize outputs from multiple AI models into a comprehensive, verified assessment. Use when: (1) User pastes feedback/analysis from multiple LLMs (Claude, GPT, Gemini, etc.) about code or a project, (2) User wants to consolidate model outputs into a single reliable document, (3) User needs conflicting model claims resolved against actual source code. This skill verifies model claims against the codebase, resolves contradictions with evidence, and produces a more reliable assessment than any single model.
Drive the NetBox Labs Platform MCP server (hosted on NetBox Cloud) from an AI agent. Covers connecting/auth, Code Mode (the three tools netbox_search_schema / netbox_execute / rediscover_netbox, the injected helper catalog, the sandbox contract, round-trip minimization) and Discrete per-tool mode. Use when an agent connects to the Platform MCP server, writes code-mode programs against NetBox, or chooses between code and discrete mode.
Operate Orca-managed worktrees, folder contexts, terminals, repos, automations, artifacts, skill sharing, worktree comments, and Orca's embedded browser through the `orca` CLI. Use when the user says "$orca-cli", "Orca worktree", "child worktree", "spawn codex/claude in a worktree", "read/wait/send Orca terminal", "handoff" / "handover" / "give this to another agent", "Orca browser", "orca artifacts", or "share skills". Prefer it over raw git worktree, ad hoc PTYs, or Computer Use when Orca state is involved. Use Computer Use only when a visible window needs GUI control that a CLI, filesystem, or API cannot do.
Turn the main session into a dispatcher that never blocks. Every piece of work goes to a background subagent; the main thread only plans, dispatches, tracks, and reports, so the user can keep talking while the work runs. Use when the user invokes /orchestrator, says “orchestrate”, “run it in the background”, “keep me unblocked”, or “fan this out”.
Create or hire Paperclip agents using the native governance-aware workflow. Use when the user asks to create, hire, provision, draft, or configure a new Paperclip agent, including choosing role, reporting line, adapter config, desired skills, instructions bundle, budget, heartbeat settings, and approval handling.
Guide for running multiple parallel Claude Code sessions using cw (Claude Worktree Manager). Use when the user wants to parallelize work across multiple Claude instances, manage git worktrees for concurrent AI coding tasks, or merge results back together. Triggers on tasks involving parallel Claude sessions, worktree management, or splitting work across multiple agents.
Explore a large codebase in parallel by launching multiple explore subagents that each investigate a different area simultaneously. Use when onboarding onto a new project, understanding architecture, or investigating a cross-cutting concern.
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.