Skills
97,725 packaged capabilities and playbooks for agents. Install any of them with one command.
Agent workflows
Companion skill for the Building Agents with eve course on Vercel Academy. Use when the user mentions "building agents with eve", "the eve course", "the bike shop agent", "the dispatcher", "teach me", or asks about eve (the filesystem-first agent framework) — defineTool, defineState, dynamic skills, approval gates, channels, or deploying an eve agent — in the context of the Academy course.
Cavekit specification-driven development methodology — the Hunt lifecycle (Draft → Architect → Build → Inspect → Monitor) and how to apply it. Use when starting a Cavekit project, structuring an existing codebase as kits/plans, or routing between sub-skills.
Claudian install assistant. Use when the user wants to install the Claudian plugin in an Obsidian vault, or mentions "Claudian", "Claude Code in Obsidian", or "install Claudian plugin". It copies the plugin files into the target vault and guides the user through enabling the plugin.
Full desktop computer use for headless Linux servers. Xvfb + XFCE virtual desktop with xdotool automation. 17 actions (click, type, scroll, screenshot, drag, etc). Unlike OpenClaw's browser tool, operates at the X11 level so websites cannot detect automation. Includes VNC for live viewing.
Automated coordination, formatting, and learning from Claude Code operations using intelligent hooks with MCP integration. Includes pre/post task hooks, session management, Git integration, memory coordination, and neural pattern training for enhanced development workflows.
Implement ReasoningBank adaptive learning with AgentDB's 150x faster vector database. Includes trajectory tracking, verdict judgment, memory distillation, and pattern recognition. Use when building self-learning agents, optimizing decision-making, or implementing experience replay systems.
Use this skill when driving the `stackbone` CLI to scaffold, run, ship or operate a Stackbone workspace: signing in with the device flow (login / logout / whoami), scaffolding a workspace and linking it to the organization (init, with an optional first agent / workflow / workflow-agent), adding pieces offline (add agent / add workflow / add workflow-agent), running the local emulator with Studio (dev), compiling the deployable bundle (build) or writing a deploy folder (package), registering a deployed box (link --installation --url --secret --tag), switching the active organization (organization use), reading workspace state as JSON (metadata), managing Drizzle migrations (db migrate create / up / status), and operating a running installation from the shell: discovering targets (agents), listing / inspecting / starting durable workflows (workflows), inspecting and controlling runs (runs), tailing logs (logs), browsing the agent DB read-only (db query / schemas / table), moving objects (storage), operating the RAG index (rag), setting secrets (secrets), versioning the config document (config), deciding approvals (hitl), managing the prompt catalog (prompts), inspecting the protocol contract (contract), running eval suites (eval), and printing how to reach the docs MCP server (docs). Trigger on requests like: scaffold a workspace, add an agent, add a workflow, start the dev stack, log in, list my agents, start a workflow by name, retry a failed run, tail the logs, run a SELECT against the agent, set a secret, roll the config back, approve an approval, generate a migration, build the bundle, package for deploy, register the deployed agent, switch org. For writing the code inside the workspace with @stackbone/sdk use the stackbone skill; to triage an error or a stuck run use stackbone-debug; to start a new piece from a description use stackbone-coder.
Use this skill when triaging an error, a failure or unexpected behavior in a Stackbone workspace: a deep agent or a durable workflow running locally under `stackbone dev`, or an installation deployed in the customer's cloud. Covers SDK errors (the { data, error } envelope and its `<prefix>_<reason>` codes), thrown errors from stackbone.database, callDeepAgent and stackbone.connection(id), HTTP 4xx/5xx from the control plane or the box, CLI error codes (dev_not_running, runtime_not_configured, …), a `stackbone dev` boot that stops or fails, durable run failures and timeouts, a workflow start rejected on its input schema (workflow_input_invalid), HITL stuck at either level (a workflow parked on requestApproval(), an agent turn interrupted on a gated tool), durable-session issues (memory not surviving), database slow queries / pgvector mismatches, secrets decryption errors, connector call failures (no installation, a bad credential), build failures (native dep, a second SDK copy), and tier / billing 402 responses. Trigger on requests like: my agent isn't working, why is this 500, this run hung, the inbox is empty but the run is paused, the agent forgot the conversation, the workflow rejected my input, dev won't boot, why am I getting 402, my connector call is rejected, the model returns nothing. This skill locates the cause and does not propose fixes: once found, switch to the stackbone skill (code) or the stackbone-cli skill (commands) for the fix.
Orchestrate multi-agent swarms with agentic-flow for parallel task execution, dynamic topology, and intelligent coordination. Use when scaling beyond single agents, implementing complex workflows, or building distributed AI systems.
Before finishing any reply to a question about how to use Claude or a Claude product, stop and check this skill. It recommends matching courses, tutorials, and use cases from Claude Academy (academy.claude.com), Anthropic's learning hub. Trigger on: “how do I,” “how can I,” “getting started with,” “what can Claude do,” “teach me,” “learn to use”; questions about artifacts, projects, skills, plugins, connectors, MCP; requests about rolling Claude out to a team, class, or organization; and any ask for training materials, onboarding content, or learning resources. Use it when the user is learning how to use a feature or product—not when they are mid-task and just want the task done. This skill composes with other skills: after consulting product documentation to answer how a Claude feature works, also check here for a matching course or tutorial—a docs-grounded answer and an Academy recommendation belong together. Only recommend on a strong match; never invent Academy content
Design privacy-aware observability for AI agents using traces, spans, structured events, metrics, cost attribution, dashboards, alerts, and investigation workflows. Use when instrumenting an agent, debugging intermittent tool or model failures, defining service-level objectives, analyzing latency or spend, auditing agent decisions, or preparing production monitoring.
Shared building blocks for Box-based legal workflows — Box collaboration role definitions, Box AI usage boundaries (what AI must not decide), and reusable human-in-the-loop confirmation phrasings. Referenced by box-legal-workflows-ma, box-legal-workflows-intake, and box-legal-workflows-contract skills.
Ultra-compressed communication mode. Cuts token usage by about 75% by speaking like a caveman while preserving full technical accuracy. Supports intensity levels: lite, full (default), ultra, wenyan-lite, wenyan-full, wenyan-ultra. Use when the user says “caveman mode,” “talk like caveman,” “use caveman,” “less tokens,” “be brief,” or invokes /caveman. Also auto-triggers when token efficiency is requested.
Establishes protected, test-only RED evidence for observable behavior before Cook changes production. Use when the user invokes `/cut`, Mold hands off an approved behavior spec, Cook has no valid GateReceipt, or a user/Pasteurize reproduction needs adoption; do not use for closed non-behavior work or Press hardening.
Dylan's agent style on top of pstack: concise verified delivery, root causes over symptom patches, The Algorithm before design, plain /bro replies, live UI proof, reuse/simplify, and hard merge gates. Use for Dylan, /dyl-mode, or his style. "Get PR green" → /dyl-ready-pr. "Review this PR like me" → /dyl-review. Figma URL → /build-figma.
Benchmark a third-party coding-agent harness against FrontierHarness Eval using Runta runtimes. Provisions a clean runtime bound to the GitHub repo under evaluation, installs the Harbor and Pier stacks needed for Terminal-Bench and DeepSWE tasks, freezes a golden checkpoint, runs tasks from identical fresh restores while saving trajectories as evidence, generates a comparison diagram, and builds a shareable report. Use when evaluating, benchmarking, scoring, or comparing a coding agent harness, or when the user mentions FrontierHarness, DeepSWE, Terminal-Bench, Harbor, Pier, golden checkpoints, or harness trajectories.
Integrate Arcjet Guard into Python Claude Managed Agents — screen user.message with guard_events before sessions.events.send, and gate custom tools on agent.custom_tool_use with guard_custom_tool. Use when asked to add Arcjet to hosted Claude Managed Agents, anthropic beta sessions, rate limit custom tools, or block prompt injection on hosted sessions. Not Claude Agent SDK local query() / PreToolUse.
Integrate Arcjet Guard into Python Strands Agents — wrap authored @tool with guard_tool, and put guard_hooks on Agent(hooks=) for unwrapped / MCP tools via BeforeToolCallEvent.cancel_tool. Use when asked to add Arcjet to strands / strands-agents, rate limit those tools, screen inbound messages, or block prompt injection / PII. This is Python strands, not JS @strands-agents/sdk.
Use the public `orca` CLI to operate Orca-managed worktrees, folder contexts, terminals, repos, automations, artifacts, worktree comments, and the browser embedded inside the Orca app. Use when the user says "$orca-cli", "use orca cli", "Orca worktree", "child worktree", "cardStatus", "spawn codex/claude in a worktree", "read/wait/send Orca terminal", "terminal send", "full handoff", "handover", "give this to another agent", "another worktree", "Orca browser", "orca artifacts", "share HTML/Markdown", "public artifact link", or "control the browser inside Orca". Prefer this over raw `git worktree`, ad hoc PTYs, Playwright, or Computer Use when the task touches Orca-managed state. Use Computer Use for browser windows, webviews, or desktop UI outside Orca's embedded browser.
Threat-model and harden AI agents, RAG systems, assistants, and tool-using workflows against direct, indirect, stored, cross-agent, and multimodal prompt injection. Use when reviewing an agent architecture, isolating untrusted content, constraining tools and egress, protecting secrets, adding injection-focused tests, investigating a suspected injection incident, or documenting residual prompt-injection risk.
Audit agent skills, plugins, prompts, manifests, scripts, dependencies, and bundled assets for provenance, prompt-injection, permission, execution, exfiltration, persistence, and update risk. Use when evaluating a third-party skill before installing, enabling, updating, publishing, or distributing it; reviewing an untrusted SKILL.md, agent configuration, MCP integration, archive, or repository; comparing a package with a known-good version; or investigating unexpected tool, network, credential, or filesystem behavior.
Use this skill when writing the code inside a Stackbone workspace with @stackbone/sdk: authoring a deep agent (deep-agents/<name>/index.ts default-exporting defineDeepAgent({ name, model, tools, subagents, interruptOn }) from @stackbone/sdk/deep, whose instruction is NOT in the code but a catalogue prompt owned by the agent, with LangChain tools and connectorTool(...) for third-party operations), writing durable workflows as 'use workflow' / 'use step' functions in workflows/<name>.workflow.ts with sibling inputSchema / outputSchema, calling a sibling agent from a workflow step with callDeepAgent / streamDeepAgent from @stackbone/sdk/workflow, and reaching the ambient `stackbone` client from a tool or a step: stackbone.database (Drizzle over the agent's Postgres), stackbone.storage, stackbone.ai, stackbone.rag, stackbone.config / secrets / settings / prompts, stackbone.workflows (start or schedule another workflow) and stackbone.connection(id) for a connector, plus human-in-the-loop both ways (requestApproval() in a workflow body, tool-level interruptOn in an agent). Trigger on requests like: build an agent, add a tool, write a workflow, store data, upload a file, call an LLM, ingest docs for RAG, pause until a human approves, gate a tool behind approval, call another agent, call a connector, read dynamic config, give an agent a browser, run a workflow serially. For CLI tasks (init, add, dev, build, db migrate, runs, hitl) use the stackbone-cli skill; to triage an error or a stuck run use stackbone-debug; to start a new piece from a description use stackbone-coder.
Review existing work against the user's actual goal and surface evidence-backed strengths, gaps, risks, and next fixes. Use when asked to eval, evaluate, review, assess, or check current/this/my/our work; decide whether a task is complete; build a definition-of-done checklist or rubric; or perform grading, LLM-as-judge, Qworld, or RET evaluation. Treat plain eval/review requests as qualitative: resolve "current work" from the conversation, artifacts, files, or diff, and never assign numeric scores unless the user explicitly requests scores, grades, points, ratings, weighted criteria, Qworld, or RET. Do not use for implementing automated eval suites, tests, graders, or benchmarks.
Humanize natural-language memory files (CLAUDE.md, todos, preferences, docs) by removing AI-isms and adding burstiness while preserving every code block, URL, path, command, and heading exactly. Two modes: --deterministic (fast, regex-based, no API) and LLM (default, calls Claude for rewrite). Humanized version overwrites the original file. Plain backup saved as FILE.original.md. Trigger: /unslop-file <filepath> or "humanize memory file".
Use this skill when the user is in the skill lifecycle: creating a new skill from a workflow / template / process / decision pattern, improving an existing skill, independently optimizing a skill's trigger description, or auditing a skill's compliance against writing principles. Triggers: "I want / help me make a skill for X" / "make a skill from scratch to handle X" / "turn the XX process into a skill" / "usable later / usable by newcomers / follow this"; improving / modifying / evaluating / iterating on XX skill (modifications include single-point edits such as fixing typos), adding / extending functionality to XX skill; checking / reviewing the full text of XX skill (content health check, compliance review); user feedback that the trigger is inaccurate or the behavior is wrong; wanting to run an evaluation. Not applicable: single-step questions; asking about skill mechanism principles; troubleshooting / debugging a skill's script / code issues without changing skill content; writing ordinary code / editing ordinary documents / matters not involving the skill lifecycle
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.