Skills
97,222 packaged capabilities and playbooks for agents. Install any of them with one command.
Agent workflows
Guidance for discovering, installing, generating, validating, and maintaining package-owned Agent Skills. Use when a project uses TanStack Intent or when the task needs its setup, implementation, debugging, testing, migration, or integration guidance.
Unified team skill for UI design team. Research -> design tokens -> audit -> implementation. Uses team-worker agent architecture with roles/ for domain logic. Coordinator orchestrates dual-track pipeline with GC loops and sync points. Triggers on "team ui design", "ui design team".
Creates and registers templates for agents, skills, workflows, hooks, and code patterns. Handles post-creation catalog updates, consuming skill integration, and README registration. Use when creating new template types or standardizing patterns.
Store and retrieve customer context using Twilio Conversation Memory. Covers Memory Store provisioning, profile management, traits, observations, conversation summaries, and semantic Recall. Use this skill to give AI agents or human agents persistent memory of customer interactions across sessions and channels.
Initialize, diagnose, or migrate a project into the LLM wiki pattern with AGENTS/CLAUDE/ANTIGRAVITY instructions, QMD MCP wiring, Claude/Codex/OpenCode/Antigravity hooks/plugins, guardrails, and QMD doctor checks. Use when the user asks to set up wiki infrastructure, check if a project needs migration, install wiki hooks, or validate QMD.
Inspect, drive, and verify a running Windows app UI via winapp ui (UI Automation): find elements, click, type, read text/values, wait for state, take screenshots in WinUI, WPF, WinForms, Win32, or Electron apps. Use to check, test, or act on a live app; for WinUI XAML internals, use winui-devtools.
WokiToki (`toki`) — a local, browser-based human-in-the-loop feedback CLI the AI drives mid-conversation to collect structured, anchored, point-by-point answers. WHEN to use: the AI needs the user to react to MORE THAN ~3 decision points at once, OR to a long multi-section explanation/report the user may want to answer one point at a time. It beats `AskUserQuestion` (which is capped at ~4 questions × ~4 options, terminal-bound, has no rich free-text, and cannot show the reference content while answering) and it beats an inline prose questionnaire (which produces unanchored replies the AI must guess-map back to each question/paragraph). HOW it works: the AI writes a spec JSON of `blocks` (a block WITH `controls` is a question, a block WITHOUT `controls` is a report paragraph, and a mix is a hybrid form), runs `toki <specPath>` (a global binary the skill builds on first use), and the user answers in a local dark-themed web UI with single/multi/toggle controls + a per-block free-text field + highlight-to-quote; on submit the tool prints ONLY the Result JSON to stdout (banners/errors go to stderr), which the AI reads the SAME turn and continues with precise, anchored feedback. Triggers on: `wokitoki`, `toki`, `feedback UI`, `ask me point by point`, `let me answer in the browser`, `decision form`, `interactive feedback`, `review this plan point by point`, `react to each section`, `more than three questions`, `long explanation I want to respond to piece by piece`. Use this skill even when the user does not say `toki` — if the task is to collect granular anchored feedback on many decisions or a long report, this is the right tool. Do NOT use for: a trivial 1–2 option pick or a single yes/no (use `AskUserQuestion`), or any non-interactive / CI / one-shot output where there is no human at a browser to answer.
Standard feature workflow with timestamped specs under `specs/`, checklist-driven plans, autonomous checkpoints, and a post-implementation review step. Use when starting a multi-file feature, when the user says "start a new feature", "how should I plan this", "what's the process", "write a spec for this", or before any non-trivial implementation that needs a planning doc.
Guide for adding new LLM models to Letta Code. Use when the user wants to add support for a new model, needs to know valid model handles, or wants to update the model configuration. Covers models.json configuration, CI test matrix, and handle validation.
Give an AG2 `Agent` the ability to run shell commands. Covers `SandboxShellTool` (client-side `subprocess` via `LocalEnvironment`, works with any provider) and the provider-native `ShellTool` (OpenAI Responses execution). Use when the user wants the Agent to execute commands, build/test code, manage files, or operate on a workspace. Always pair with sandboxing — `allowed`, `blocked`, `ignore`, or `readonly`.
Add OpenTelemetry traces to an AG2 `Agent` via `TelemetryMiddleware` (`ag2.middleware.builtin`). Emits spans for the full turn, each LLM call, each tool execution, and each human-input request, following the OpenTelemetry GenAI semantic conventions. Compatible with any OTLP backend — Jaeger, Grafana Tempo, Datadog, Honeycomb, Langfuse. Use when the user wants production-grade traces, latency analysis, token-usage attribution, or to ship telemetry into an existing observability stack.
Use when the work spans several files or replaces a framework, language, auth, storage, or API layer (rewrites and migrations included), and the user wants it to end up working and tested. Asks only what is missing, writes a spec with acceptance criteria, then runs and verifies it in the background. Not for questions, small edits, or an existing Seed (use run)
Delegate a bounded task to the locally authenticated Google Antigravity CLI (`agy`) and return a host-verified handoff. Use when the user explicitly asks to call Antigravity, `agy`, or `$call-agy` for coding, review, analysis, testing, or verification. Exclude informational questions and generic subagent requests.
Ultra-compressed communication mode. Cuts token usage ~75% by dropping filler, articles, and pleasantries while keeping full technical accuracy. Use when user says "caveman mode", "talk like caveman", "use caveman", "less tokens", "be brief", or invokes /caveman.
Report ground-truth Claude token consumption and estimated cost by parsing JSONL session files directly. Use when checking API spend, auditing token usage by project/session/model, generating daily/weekly/monthly cost reports, or diagnosing ccusage undercounting. Includes subagent files that ccusage ignores.
This skill should be used when the user asks to "create a team", "spawn agents", "assign tasks", "coordinate multiple agents", "check team status", "view kanban board", "send messages between agents", "manage team tasks", "monitor team progress", or mentions "oh", "multi-agent coordination", "team collaboration", "agent inbox", "task board", "spawn worker". This skill should also be triggered when the current task is complex enough to benefit from splitting into subtasks and delegating to multiple agents — for example when the user asks to "build a full-stack app", "refactor the entire codebase", "implement multiple features in parallel", or when the agent determines that the work scope exceeds what a single agent can efficiently handle alone. Provides comprehensive guidance for using the ClawTeam CLI to orchestrate multi-agent teams with task management, messaging, monitoring, runtime profiles, git context, and recovery tooling.
Read project config from .agents.yml and .agents.local.yml with local overrides. Supports dot notation for nested fields. Invoke when agents/skills reference `config_read("field", "default")` pseudocode. Args format "FIELD DEFAULT".
Create new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, edit, 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
Confirmation gate before any bulk delete, cleanup, or destructive operation that could result in data loss — shell-level (rm -rf, git rm, bulk sed) or brain-level (bulk forget, delete sweeps, purge-deleted, source removal, raw-SQL truncation). Presents a recoverability card and requires an explicit "yes" from the user before proceeding. Routing convention, not an operation-boundary enforcement.
Scans for project documentation files (AGENTS.md, CLAUDE.md, GEMINI.md, COPILOT.md, CURSOR.md, WARP.md, and 15+ other formats) and synthesizes guidance. Auto-activates when user asks to review, understand, or explore a codebase, when starting work in a new project, when asking about conventions or agents, or when documentation context would help. Can consolidate multiple platform docs into unified AGENTS.md.
Set up Claude Code hooks to block dangerous git commands (push, reset --hard, clean, branch -D, etc.) before they execute. Use when user wants to prevent destructive git operations, add git safety hooks, or block git push/reset in Claude Code.
Add, audit, or fix GitHub Copilot repository custom instructions — `.github/copilot-instructions.md` (repo-wide) and `.github/instructions/*.instructions.md` (path-specific) — so Copilot Chat, Copilot code review, and the Copilot coding agent all pick up project-wide guidance. Also covers `AGENTS.md`, `CLAUDE.md`, and `GEMINI.md` as alternate agent instruction files. Use when the user says "add Copilot instructions", "set up copilot-instructions.md", "audit our Copilot config", "make Copilot follow our conventions", "configure Copilot code review", "add path-specific Copilot rules", "applyTo glob", "excludeAgent", "Copilot is ignoring our guidance", or invokes /github-copilot-repo-instructions. Do NOT use for personal (per-user) Copilot instructions; those are out of scope for this skill. For AGENTS.md authoring guidance beyond file layout, see the agents-md skill.
Use when generating, reviewing, or improving Hexabot v3 workflow YAML, workflow actions, bindings, memory definitions, or agentic automation use cases, including conversational, manual, scheduled, event-driven, system, sales-demo, documentation, and product-showcase workflows.
Route Data Analytics plugin-level requests and broad analytics work to the right focused workflow. Use when Data Analytics is at-mentioned, or for analytics requests involving data, metrics, dashboards, reports, charts, notebooks, spreadsheets, KPIs, market sizing, or semantic layers.
Triggers when the user says 「批量修复」「批量 jira-fix」「jira-fix-batch」「批量修复多个 Jira」「批量修复以下 bug」/ batch fix, batch jira-fix. Thin trigger shell: parse Jira IDs/URLs and enqueue them into goal-driven-batch with Engine jira-fix-workflow. Does not run fixes or merge. Do NOT use for a single issue (jira-fix-workflow) or a non-Jira queue (goal-driven-batch).
Use when executing an approved Krypton plan, GOAL.md, or implementation plan that already defines intent, ownership, contract, cutover, task boundaries, and acceptance evidence. Use for main-agent execution with explorer, plan-reviewer, reviewer, maintainer, or verifier gates.
Local memory management for agents. Compression detection, auto-snapshots, and semantic search. Use when agents need to detect compression risk before memory loss, save context snapshots, search historical memories, or track memory usage patterns. Never lose context again.
Guidance for managing AI agents at enterprise scale with Microsoft Agent 365 - the control plane that lets admins observe, govern, and secure every agent (Microsoft, Copilot Studio, and third-party) from a single registry, using Microsoft Entra Agent ID for identity, Microsoft Purview for data security, and Microsoft Defender for runtime threat protection. WHEN: Microsoft Agent 365, Agent 365, manage agents at scale, agent control plane, agent registry, agent map, observe govern secure agents, Entra Agent ID, agent identity lifecycle, third-party agent governance, secure AI agents tenant-wide, agent sprawl, agent inventory. DO NOT USE for Purview-only data controls on a single agent (use purview-agent-365-security) or end-user Copilot oversharing (use purview-copilot-oversharing).
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.