Skills
96,770 packaged capabilities and playbooks for agents. Install any of them with one command.
Agent workflows
Describe what this skill does and when an AI agent should use it. Be specific — this field determines whether Claude loads this skill. Bad: "Analytics tool for blockchain data" Good: "Use when the user asks to analyze on-chain token metrics, wallet PnL, or DEX volume using Dune Analytics queries"
Execute an external Pi CLI prompt with a provider and model pinned per call. Use only for an explicit Pi CLI request, a selected Pi provider route unavailable as native host delegation, or an approved external fallback, including Kimi, GLM, Qwen, Gemma, Muse, Minimax, and Mistral seats.
Guide plugin development workflow — editing skills, agents, hooks, or eval framework in this repo. Use when modifying files in plugins/elixir-phoenix/, lab/eval/, or lab/autoresearch/. Ensures changes pass eval, lint, and tests before committing.
Build Retrieval-Augmented Generation (RAG) systems for LLM applications with vector databases and semantic search. Use when implementing knowledge-grounded AI, building document Q&A systems, or integrating LLMs with external knowledge bases.
Reliably show images, renders, mockups, screenshots, or plots to the human user inside the Cursor app chat. Use whenever the user must SEE an image the agent generated or found, or when the user says they cannot see an image you sent. Differentiator: fixes the silent failure where images saved outside the open workspace never render in Cursor chat.
Requirement design workflow - mandatory before any creative development (new features, components, behavior changes, API/DB design): triage, parallel exploration, one-question-at-a-time clarification, adversarial validation and 2-3 option comparison produce a spec, then hand off to writing-plans. Use exploring first while the idea is unsettled; not for pure Q&A, test runs, or no-design-space small fixes (use quick-fix).
Use when the user asks to take a bug, note, defect, improvement, or task from Scout and handle it like a professional AI operator working from Scout. Also use for short commands like "do the next Scout task", "сделай следующую задачу из Скаута", or "возьми задачу из Scout".
Master skill — set up the full agent harness for any repo so an agent can work it reliably: legible (map-not-manual docs + custom lints), executable (one-command dev stack), verifiable (e2e gate + a verify-before-ship loop), plus commit hygiene and entropy control. Use when onboarding a new/unfamiliar codebase to agent-driven development — "set up the harness", "make this repo agent-ready", "harness this codebase".
Create, set up, or update the personal context portfolio: structured markdown files describing who you are, how you work, your teams, and your tool/ADO configuration. Runs the interview workflow for first-time setup and targeted edits for updates. Trigger this skill when the user asks to: set up their context, create or update their context portfolio, "create my IQ", "set up my IQ", edit their profile, add/remove a stakeholder, update ADO config, change team info, update pillars, or set up any plugin configuration. Trigger when another skill fails to find context (missing files or TODO markers) and needs context populated. Also trigger when the user mentions a context change in passing (e.g., "my manager changed", "we added someone to the team") to offer a context file update. Do NOT trigger for read-only questions like "who's on my team?" or "what's my ADO config?". Those are answered directly from the context files referenced in the loaded custom instructions; no skill is needed.
Execute an accepted engineering plan through implementation, integrated validation, code review, pull request creation, active review monitoring, and verified closeout. Use after Kickoff approval/review or when the user explicitly asks to ship an existing plan.
Used to hide or restore skills; supports single-skill or batch operations; manages skill parent folder paths; prevents hiding itself; supports multi-directory search and operating on a specified directory; supports one-click hide/restore of all skills. Use this skill when the user needs to [hide or restore skills].
Use this skill to audit, review, or validate Claude Code skills (.md files in .claude/commands/). Invoke when user wants to check skill quality, cross-platform compatibility, cross-agent compatibility, prerequisite declarations, or description triggering accuracy. Triggers on: 审查 skill, review skill, 检查 skill 质量, skill 兼容性检查, validate skill, audit skill. This skill delegates structure validation to validate.sh, content quality to skill-creator, and handles compatibility auditing itself. Do NOT use for general code review, reviewing PRs, or reviewing CLAUDE.md.
Unified task execution protocol for Codex-only work. Supports Single Task, Epic Task, and Batch Task while preserving CSV truth-source, validation gates, context recovery, and Debug-First failure exposure. WHEN TO USE: user asks to "track tasks", "create todo list", "make a plan", "track progress", "long task", "big project", "build from scratch", "autonomous session", "跟踪任务", "自主执行", "长时任务", "从零开始", "任务管理", "做个计划", "大工程", or when a task clearly requires 3+ ordered steps that produce file changes. DO NOT USE: single-step fixes, pure Q&A, code review, explaining code, search/research tasks, tasks with fewer than 3 steps, or tasks that do not produce file changes.
Stretch Claude Code context. Use when sessions are running long, hitting context/usage limits, working through large repos, doing multi-day work, or burning through tokens faster than expected. Apply these habits proactively on any non-trivial task.
Use when design is complete and you need detailed implementation tasks for engineers with zero codebase context - creates comprehensive implementation plans with exact file paths, complete code examples, and verification steps assuming engineer has minimal domain knowledge
End-to-end implementation and self-review in a single invocation — implements from an Issue, plan file, or free-text description, then runs a lean fresh-context review. With --guard it also waits for CI and peer review, addresses feedback until the PR is review-ready, and ends with a risk verdict (merge-ready or review-ready, or merged when the project opts in). Use when the user wants to implement and review without manual handoff between skills
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 "$orca-cli", "Orca worktree", "child worktree", "spawn codex/claude in a worktree", "read/wait/send Orca terminal", "handoff" / "handover" / "give this to another agent", "Orca browser", "orca artifacts", or "share skills". Prefer it over raw git worktree, ad hoc PTYs, or Computer Use when Orca state is involved. Use Computer Use only when a visible window needs GUI control that a CLI, filesystem, or API cannot do.
Evaluate changes to agent prompts, skills, routing policies, and harnesses against reproducible baselines, held-out suites, and regression benchmarks. Use when optimizing agent system prompts, measuring skill triggering accuracy, evaluating routing changes, or running benchmark regressions — even if the user does not explicitly say "fable-eval" (e.g. "benchmark this prompt", "evaluate agent behavior", "test skill performance", "run the eval suite"). Do NOT use for routine application unit tests (use fable-verify).
Author, evaluate, refine, optimize, and package autonomous AI agent skills across multi-agent ecosystems with BinEval scoring and description tuning. Use when creating a new skill from scratch, editing an existing skill that misfires or undertriggers, authoring test suites for skills, optimizing skill descriptions, or packaging skills for distribution — even if the user does not explicitly say "fable-skill-creator" (e.g. "create a skill", "build a new skill", "optimize skill description", "package this skill", "teach the agent to do X"). Do NOT use for general application code changes or non-skill tasks.
Operate Orca-managed worktrees, folder contexts, terminals, repos, automations, artifacts, skill sharing, worktree comments, and Orca's embedded browser through the `orca` CLI. Use when the user says "$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.
Use when the user wants to audit or correct drift across canonical project context files such as `AGENTS.md`, `CLAUDE.md`, `ARCHITECTURE.md`, `CONSTRAINTS.md`, `JARGON.md`, `EPISTEMIC-MAP.md`, `README.md`, and `openspec/config.yaml`, or mentions context drift, memory-file drift, redundant project docs, wrong document, layered documentation, or doc ownership. This skill checks whether each fact lives in the right document, identifies redundancy and layer-boundary violations, and prepares lossless move/delete proposals with diff previews and per-change approval. Prefer it over ad-hoc cleanup because these files are load-bearing and misplaced content can change agent behavior or tool behavior.
Act as the Light workflow advisor, navigator, and router. Understand what the user is trying to do, inspect the actual project/host evidence and Skill catalog, reason about the best next Light Skill, explain why, wait for user approval, then honor the accepted Skill's invocation policy with host-aware transition. Also answer collection-navigation and standalone routing requests. Use only when the user explicitly invokes $ask-light; never execute before consent
Behavioral guidelines to reduce common LLM coding mistakes. Use when writing, reviewing, or refactoring code to avoid overcomplication, make surgical changes, surface assumptions, define verifiable success criteria, keep a WORKLOG.md so work can be continued across sessions, and write research to research/ and plans for review to docs/.
Turn conversations, transcripts, project notes, folder workflows, documentation, or other sources into reusable agent skills and Skill Creator compatible instructions. Use when a user asks to create, update, distill, or "learn" a workflow as a skill; preserve repeated operating methods for future AI agents; encode corrections/failure modes as durable procedure; or transform source material into portable skill guidance for coding agents, research agents, automation agents, or custom agent frameworks
Use to build, manage, and improve the context that makes Hex's AI agents (Threads, the Notebook Agent, the Modeling Agent) accurate — endorsements & exclusions, warehouse descriptions, workspace context & guides, and semantic models. Covers standing context up from scratch, measuring it with a dedicated eval engine (authoring and running eval suites, sourcing cases from real usage, gating changes in a preview), reviewing Context Studio Suggestions, and publishing via GitHub sync or the Hex CLI. Triggers on: “context management”, “manage my Hex context”, “context strategy”, “context engineering”, “set up Threads context”, “workspace guide”, “warehouse descriptions”, “endorse tables”, “semantic model”, “my agent gave the wrong answer”, “audit my Hex context”, “improve agent accuracy”, “eval my context”, “write Hex evals”, “eval rubric”, “baseline vs hill-climbing”, “convert my evals”, “test context changes”, “hex context preview”, “context suggestions”, “Hex Context Studio”. Use for any task about making Hex's agents trustworthy through better context—even phrased casually like “help me get my data team using AI” or “why does Threads keep picking the wrong table”. Routes to the specific workflow rather than running the full setup every time.
Run GenAI evaluations on SAP AI Core: benchmark LLMs and prompts against a test dataset using system-defined and custom (LLM-as-judge) metrics, then retrieve aggregate results. Supports all three evaluation types — prompt registry, orchestration registry, and grounding (RAG) orchestration registry. WHEN: user wants to evaluate or benchmark an LLM / prompt / orchestration config, run a GenAI evaluation, create or list evaluation metrics, define a custom LLM-as-a-judge metric, create a prompt template (prompt registry) or an orchestration-registry config (with or without grounding) for evaluation, create an evaluation configuration or execution, or fetch evaluation results / scores for an execution. DO NOT USE FOR: agent evaluation via the offline aeval service — use `aeval-offline-evaluation`; building a production orchestration/RAG pipeline for inference — use `genai-hub-orchestration`; generic executions, deployments, or artifacts unrelated to evaluation — use `aicore-lifecycle-management`; credentials, resource groups, or object stores — use `aicore-admin-resources`.
Configure and use the SAP AI Core Orchestration service: build prompt templates, add content filtering, set up document grounding (RAG) pipelines, and call models via the Harmonized API. WHEN: user asks about the orchestration service, Harmonized API, prompt templating with variables, content filtering, grounding, RAG pipeline, document grounding, or building a pipeline that wraps an LLM. DO NOT USE FOR: creating an orchestration deployment — use `aicore-lifecycle-management` for that; credential setup — use `aicore-admin-resources`; browsing or deploying foundation models directly without a pipeline — use `genai-hub-foundation-models`; direct LLM inference without a pipeline — use `aicore-sdk-java` or `aicore-sdk-js`
Set up Git, AGENTS.md, CLAUDE.md, and the initial commit point for a new project in a practically empty folder. Use for requests such as “set up the project,” “create a new project,” “prepare the workspace for a new project,” or “initialize this empty folder.” Do not use in folders that already contain code or project configuration.
Designs a Replay Vision scanner that produces trustworthy observations: one visible question per scanner, a type chosen from the answer shape, a query that selects only sessions able to answer it, a prompt that demands on-screen proof and allows no or inconclusive, a model picked by the cost of a wrong answer, and a first-batch test pass. Sizing and the create call stay in creating-replay-vision-scanners. TRIGGER when: user wants to design, draft, or improve a Replay Vision scanner or its prompt, asks what to scan for, asks why a scanner returns vague observations, or asks an agent to propose scanners for a product. DO NOT TRIGGER when: the question and prompt are settled and the job is to size and create (use creating-replay-vision-scanners), the scanner targets one experiment (use scanning-experiments-with-replay-vision), or the job is to read existing observations (use exploring-replay-vision-observations).
Plan a large, multi-session, decision-heavy effort as a persistent decision map of tickets. Use only when the user explicitly invokes $decision-map; it maintains .scratch/<effort>/map.md and child decision tickets until fog clears, then hands off to project-spec
Walk a person from an idea to a Claude Managed Agents agent that runs on its own, and stay with them through debugging. Explains each concept at the moment it matters (Agent, Environment, Session, Deployment, vault, memory store, outcome), asks the real decisions as illustrated AskUserQuestion choices, designs the system prompt in Simplified Technical English, finds MCP connectors, sets up credentials without the secret ever reaching the chat, writes the files, applies them with the `ant` CLI, runs a capped test session, and diagnoses failed sessions from their events. Trigger on "crea un agente en Managed Agents", "managed agents wizard", "quiero un agente que corra solo", "agente programado", "agente always-on", "agente que revise mis repos cada lunes", "mi sesión se quedó pegada", "por qué falló mi agente", "debug my managed agent session", "set up a scheduled Claude agent", "deploy an agent on a cron". Not for the Claude Agent SDK (use agent-sdk-wizard), the raw Messages API, or Claude Code itself.
Clarify a real project's unresolved decisions from inspected project facts and return a bounded handoff artifact. Use only when the user explicitly invokes $project-clarify after readiness or when a project needs formal clarification; it does not create a SPEC, tickets, or automatically start another user-invoked stage
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.