Skills
97,725 packaged capabilities and playbooks for agents. Install any of them with one command.
Agent workflows
Personalized audit that teaches advanced Superset features the user isn't using yet — automations, parallel agents, tasks, multi-host, terminals, custom commands, MCP. Use when the user wants to get more out of Superset, learn advanced Superset features, or 10x their Superset workflow.
Install, set up, verify, update, pin, uninstall, or troubleshoot the ToolUniverse plugin on OpenAI Codex. ALWAYS consult this skill for any of those — don't answer from memory, because the exact marketplace name (mims-harvard/ToolUniverse), the "codex plugin marketplace add" then "codex plugin add -m tooluniverse" flow, Codex's startup auto-upgrade behavior, the uvx tooluniverse MCP server, and the API-key env vars are easy to get wrong. Use it whenever someone wants to get ToolUniverse (or "the 1000+ scientific tools" / "the harvard tools") working on Codex, says the Codex plugin or its tools/skills won't load, hits a uvx or MCP-server startup error, asks how Codex updates it, wants to pin or remove it, or finds it running an old tool version — even if they never say the word "plugin". Not for the Claude Code plugin (use tooluniverse-claude-code-plugin), for running research with the tools, or for authoring new tools or skills.
Migrates skills, agents, and rules between Google Antigravity 2.0 (IDE/CLI) format and other Agentic AI Provider formats (Cursor, Windsurf, Claude Code, GitHub Copilot, Roo Code), and vice-versa. Use when asked to migrate/convert agent customizations, rules, or skills to or from Antigravity.
Adversarial verification of finished work. Treats any "done" as a set of claims, then re-runs the claimed verifications, diffs what actually changed, detects weakened tests and false completion claims, and delivers an evidence-based verdict (VERIFIED / VERIFIED WITH CAVEATS / REFUTED). Use after any agent or model claims work is complete - "/fable-judge", "judge this work", "verify what it did", "did that actually work?". Also runs the fable-method trap suite against a skill or model via "/fable-judge suite TARGET".
Run a GEPA learning cycle via `metaharness learn` (upstream ADR-235, [email protected]) — optimizes a harness genome against a SWE-bench-style slice manifest. $0 dry-run by default; `--run` is the explicit spend opt-in. Requires a metaharness repo checkout (`--repo` or $METAHARNESS_REPO) — without one it reports `checkout-required` with clone instructions. Degrades gracefully when metaharness is absent.
Interactive feature development workflow from idea to implementation. Creates requirements (EARS format), design documents, and implementation task lists. Use when creating feature specs, requirements documents, design documents, or implementation plans. Triggered by "kiro" or references to .kiro/specs/ directory.
Connect Claude Code or Codex itself to Logfire telemetry, or route a coding agent's model requests through Pydantic AI Gateway. Use for coding-session recording, capture policies, Gateway agent configuration, temporary proxy launches, status, or removal. Not for instrumenting an application or routing an application's model client.
Guide for creating high-quality MCP (Model Context Protocol) servers that enable LLMs to interact with external services through well-designed tools. Use when building MCP servers to integrate external APIs or services, whether in Python (FastMCP) or Node/TypeScript (MCP SDK).
Semantic search over agent memory files. Use when you need to find conceptually related memory blocks, discover forgotten reference files, check what you already know before creating new memory, or search beyond exact keyword matching. Currently supports QMD (local, no API keys).
Run every code reviewer in parallel over a local diff and merge their findings into one ranked list. Use when the user asks for a review of uncommitted changes, says "quick review", or after significant code changes; and when another skill needs the local-diff review (`done` runs it as its review step).
Use this skill whenever the user is working with the Pydantic AI framework — including building AI agents, defining structured outputs with Pydantic models, wiring up tools/function calling, configuring model providers (OpenAI, Anthropic, Gemini, etc.), managing dependencies via agent context, handling streaming responses, or debugging agent runs. Trigger this skill even for adjacent tasks like "how do I make my agent return JSON", "set up a multi-step agent", "add a tool to my agent", or "validate LLM output with Pydantic" — any time Pydantic AI is mentioned or implied as the target framework.
Use when the developer asks to build or change something and the request leaves real decisions open - a new feature or flow ("build X", "lam luong Y", "them chuc nang Z"), a behaviour with more than one reasonable shape, a security or audit trade-off, or scope that could be read two ways. The request arrives as plain words, not a command - this skill fires on the words. Surface those decisions as questions with options and trade-offs, get them settled, and only then move to impact and write. Also use the moment you catch yourself choosing a business behaviour mid-implementation that the developer never stated.
Read a partner's agent codebase, score it across the 12-dimension catalog, and emit an evidence-cited markdown plus branded scored HTML report. Use for a first-touch audit: assess/review/audit our agent, where to improve, spot low-hanging fruit, an honest read of what's weak, what Ratel would notice, or `/ratel-assessment`. Static-only unless a live telemetry sample (any OTLP/OTel backend — Ratel, Langfuse, LangSmith) is MCP-wired; writes to <repo>/.ratel/ without editing code; routes into follow-up Ratel skills.
Inspect an existing agent codebase, map each agent's tools and skills, then implement Ratel so every agent discovers and loads only its intended capabilities. Use when asked to add, wire, set up, or integrate Ratel; connect existing tools or SKILL.md playbooks to an agent through Ratel; or migrate a Vercel AI SDK, Mastra, custom TypeScript, or Python agent to Ratel. Prefer Ratel's official TypeScript adapters for AI SDK and Mastra. Edits code and dependencies, preserves capability boundaries, and verifies the result. Excludes telemetry, experiments, pilots, A/B tests, and rollout planning.
Research Dream. A memory-consolidation skill for research scenarios. When accompanying a research user long-term in any host environment (OpenClaw, Codex, Claude Code, Cursor, etc.), it gradually updates USER.md, SOUL.md, MEMORY.md and memory/YYYY-MM-DD.md through everyday conversation, and periodically performs a dream-like deep consolidation so that the memory files progressively move toward a Tashan-style research digital twin.
Evaluate Agent Skills (SKILL.md directories) and answer \"is this skill any good / safe to install\" — spec compliance with security scanning, with/without behavior comparison, and description trigger testing, ending in a clear install / fix-first / reject verdict. Use this skill whenever the user asks to evaluate, audit, test, review, or score an Agent Skill or a skill collection, wants to know whether a skill actually helps or is safe, or wants to check if a skill's description triggers correctly, even if they just ask \"is this skill any good\".
Make a repository Superset-ready — author .superset/config.json with setup/teardown/run scripts so every new workspace boots configured, then verify with a real workspace. Use when the user wants to set up a project or repo for Superset, configure workspace setup scripts, or fix a failing workspace setup.
Project onboarding and session entry point for swain. On first run, performs full onboarding: migrates CLAUDE.md to AGENTS.md, verifies vendored tk, configures pre-commit security hooks, and offers swain governance rules — then writes a .swain/init.json marker. On subsequent runs, detects the marker and runs the per-session fast path (greeting, focus lane, session state). Use as a single entry point — it routes automatically. Triggers also on: 'session', 'session info', 'focus on', 'tab name'.
System maintenance with three core operations - integrity check (find/fix broken references), document session (current transcript), document recent (catch-up since last update). Plus security workflows. USE WHEN integrity check, audit system, document session, document this session, document today, document recent, catch up docs, what's undocumented, check for secrets, security scan, privacy check, OR asking about past work ("we just worked on", "remember when we").
Produces phased task boards from feature requests: dependency-mapped work items, parallelization flags, risk flags, edge cases, test matrices. Triggers on: "decompose this feature", "task breakdown with dependencies", "phased implementation plan", "work breakdown structure". NOT for discovering which decisions are still open, use decision-map. NOT for effort estimates, use estimate-calibrator.
Drive the tty7 terminal workbench from the shell with the `tty7` binary — list workspaces/tabs/panes, split a pane, send text or keystrokes into one, capture what is on a pane's screen, run a command in a real PTY and pass its exit code through, run a command in a shell pane you already have and get back its output and exit code, block until a pane finishes or needs input, see which coding agents are running and which ports a pane is listening on. Use this whenever tty7, panes, workspaces, or `%42`/`@7`/"the other pane"/"the other agent" come up; whenever you want to hand work to another agent and collect the result ("get Claude/Codex to do X", "派个活", "let another agent handle this", running several agents in parallel and merging what they produce); whenever you need to start something long-running or interactive (dev server, REPL, ssh session, `tail -f`, a TUI) that should not sit blocking your Bash tool; whenever a program needs a real terminal to behave the way the user sees it; and whenever you need to look at or report on what is running in some *other* terminal on this machine. Cheap to check: if `$TTY7_PANE` is set you are already inside tty7 and every command here works with no setup.
Use when starting any conversation that touches Oracle Database or PL/SQL - it establishes how to pick the right pythia skill BEFORE any response or action, including clarifying questions. Fires on plain requests in any language, no slash command needed - "build/add/fix/explain/review/deploy X", "lam/them/sua/tao/giai thich/kiem tra/trien khai X" (Vietnamese, with or without diacritics). Also use whenever unsure which pythia skill applies, or on catching yourself answering an Oracle question with none open.
Runs a progressive interview, without requiring UX or design vocabulary, to create or evolve a software project's canonical UX standard and generate three coordinated artifacts: UX_STANDARD.md, a custom ux-review skill, and UX_REVIEW_TEMPLATE.md. Use when you want to define look and feel, audience, principles, design system, accessibility, responsiveness, tone of voice, references, interactions, quality criteria, or UX governance before reviewing features. The generated ux-review becomes the SDD step between Define and Design. It does not run a feature review in place of the generated skill.
Scaffold and operate an opt-in `Agent activity log` table that records what the agent did, decided, and got blocked on across a long-running or multi-session Airtable workflow. Use whenever a workflow skill (product-ops, sales-ops, marketing-ops, etc.) is being set up for an agent-driven motion (recurring triage, multi-step plan, automated monitoring, agentic workflow), or when the user explicitly asks for "agent activity tracking," "audit log of agent decisions," "agent memory," "track what the agent did," or similar. The pattern is opt-in (front-load the offer, frame as auditability for the user's benefit, not surveillance). Composes into workflow skills the same way `show-airtable-link` does — workflow skills point at this skill rather than re-implementing the schema inline.
macOS platform guide for the agent-desktop CLI. Use for desktop automation on macOS: granting Accessibility and Screen Recording permission (TCC), what headed mode does on macOS (physical-first click, type and scroll), surfaces (menu, menubar, sheet, popover, alert), Notification Center, Chromium and Electron apps including launch --cdp, and fixing PERM_DENIED, empty accessibility trees, ACTION_FAILED and APP_UNRESPONSIVE.
Fixes web search on an agent whose model backend can't run it: a relay/reseller proxying Claude or Codex returns empty instead of failing. Use when web search returns nothing, a model insists a shipped product doesn't exist, someone wants to give an agent internet access, or the user is on a third-party base URL, relay, or 中转站. Diagnoses which built-in tools are dead, removes them, and installs a working replacement.
Expert in building comprehensive AI systems, integrating LLMs, RAG architectures, and autonomous agents into production applications. Use when building AI-powered features, implementing LLM integrations, designing RAG pipelines, or deploying AI systems.
Autonomous NeMo-RL research agent workflow for directed hypothesis testing and open-ended discovery. Guides agents through the full experiment lifecycle: understanding recipes and environments, wiring RL or NeMo-gym runs, launching reproducible baselines and iterations, analyzing results, preserving human oversight, and using git plus TSV logs as the research ledger. Do NOT use for: bug fixes, code review, documentation, refactoring, dependency updates, or single-file changes.
Automate browser interactions for web testing, form filling, screenshots, and data extraction. Use when the user needs to: - Navigate websites and interact with web pages - Fill forms and click buttons - Take screenshots of web content - Test web applications - Extract information from web pages - Debug frontend issues in the browser - Monitor console logs and network requests - Record browser sessions as video This skill uses agent-browser for comprehensive browser automation. RECOMMENDATION: Disable Chrome DevTools or Playwright MCP when using this skill to save context.
Given a freshly-cloned external repo/tool, run clone→explore→verify→write to produce a bundle of RUNNABLE workflow skills under .overstack/doyourmagic/<repo>/skills/ — one hub skill /dym-<repo> (1 dòng context, chỉ nạp workflow con khi được gọi) + N sub-skills dạng SKILL.md promote thẳng lên repo chính (npx skills add) — plus workflows.md (index + bảng kiểm chứng) and flow.html (docs-site-macos, sơ đồ luồng skill→sản phẩm→skill kế). Trigger on 'kéo repo mới về', 'clone tool này làm workflow', 'doyourmagic', 'onboard external tool/repo', 'generate workflow docs/skills for this repo', or /doyourmagic. Options: --name <prefix> · --humanize.
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.