Skills
97,222 packaged capabilities and playbooks for agents. Install any of them with one command.
Agent workflows
Iterates one-by-one through a multi-bug report with a hard confirmation gate before any code touches disk. Use this skill when the user has a report of 2+ bugs and wants to work through them sequentially. Trigger phrases: "let's tackle these", "let's iterate", "go one by one", "loop through the findings", "work through the bugs", "bug iteration", or any time the user presents a numbered list of bugs and signals they want to resolve them in order. Do NOT trigger for a single bug (just investigate and fix) or when the user wants a fresh debug pass (use a dedicated debug/investigation pass for that).
集中兵力：多件事同时争抢时间、注意力或预算时，列出全部待办，选定唯一主攻目标并公开锁定，彻底解决、验证后再转向下一个。当待办过多、推进分散、“每件都做了一点但都没做完”、需要决定先做什么时触发；只有一个任务、任务彼此独立可并行、用户明确要求同时推进时不触发。 English: Concentrate forces. When several tasks compete for time, attention, or budget, enumerate them, lock onto a single main target, finish and verify it completely, then move on. Trigger when priorities sprawl, effort is scattered, or you must decide what to do first; skip when there is one task, tasks are independent and parallel, or the user explicitly wants them advanced together.
Use this skill for in-depth interviews and to build a complete picture of any topic. Use it when the user says “grillme,” “ask questions,” “ask me questions,” “interrogate me,” “I want to work through a topic,” “help me figure it out through questions,” “questions?,” “I need the full picture,” “find out everything from me,” “interview me,” “ask me about it,” “grill me,” “draw it out of me,” or “roast me.” Also use it when the user describes a task superficially and you need to dig into the details before starting work.
Use the public `h0x` CLI to operate h0x-ADE-managed worktrees, folder contexts, terminals, repos, automations, artifacts, skill sharing, worktree comments, and the browser embedded inside the h0x-ADE app. Use when the user says "$orca-cli", "use h0x cli", "h0x-ADE worktree", "child worktree", "cardStatus", "spawn codex/claude in a worktree", "read/wait/send h0x-ADE terminal", "terminal send", "full handoff", "handover", "give this to another agent", "another worktree", "h0x-ADE browser", "h0x artifacts", "share HTML/Markdown", "public artifact link", "share skills", or "control the browser inside h0x-ADE". Prefer this over raw `git worktree`, ad hoc PTYs, Playwright, or Computer Use when the task touches h0x-ADE-managed state. Use Computer Use for external browser windows, webviews, or desktop UI only when the task requires OS/window-level control such as focus, menus, dialogs, coordinates, or screenshots. Use `orca-cli` for h0x-ADE's embedded pages and a page-automation tool such as Playwright or CDP for external pages.
Create a concise continuation prompt that a fresh agent session can paste in to resume a long or degraded session. Use when the user asks for a handoff prompt, restart prompt, continuation prompt, context transfer, fresh-session resume, or a compact summary for opening a new session. Do not use for ordinary summaries, task-forest maintenance, durable user-profile updates, automatic session creation, code execution, or external publishing.
Opt-in recreation of the Bun + TypeScript framework used to build Vapi's landing-page voice agents from a ROUGH_DRAFT.md spec, including a scenario registry, language stacks, prompt composer, assistant builder, and idempotent bootstrap script. Use only when the user explicitly invokes vapi-bootstrap-framework or specifically asks to reproduce or experiment with the landing-page-agent architecture. Do not use for ordinary Vapi assistant builds; use create-assistant, create-tool, create-squad, and vapi-prompt-builder instead. Targets Bun + TypeScript + @vapi-ai/server-sdk.
Iterative code refinement through plan → code → evaluate → refine cycles. Runs lint checks (ruff), tests (pytest), and structured self-evaluation each cycle, then diagnoses failures and refines. Decomposes complex tasks into sequential phases, iterating up to 3 times per phase (10 total). Use when: the main agent delegates a code task with 'MODE: MORE_EFFORT', the user selects 'More Effort' code generation mode, or the task explicitly requests iterative refinement for higher code quality. Do NOT use for single-pass code generation (Lite mode), experiment pipeline orchestration (use experiment-pipeline), or diagnosing a specific experiment failure (use experiment-craft).
Entry point for local AI conversation history across Claude Code, Codex and Kimi CLI: routes each request to the reader or continuation skill that owns it and keeps the cross-provider inventory. Use when the platform is unknown or plural ("what was I working on", "we discussed this once") and for any Kimi CLI history. When platform and action are already clear, load that skill directly.
This skill provides comprehensive guidance for using OpenCode, the open-source AI coding agent. Use this skill when working with OpenCode CLI commands, keyboard shortcuts, agents (build/plan), slash commands, tools, skills, MCP servers, or configuration. Automatically triggered when OpenCode-specific questions or tasks are detected.
Standard flow from any task source (link or description) to creating a PR: resolve task, create branch and TODO.md, delegate fix to SubAgent (TODO.md), then commit and create PR against origin only. Use find-skills to discover data-source query methods; after confirmation save to global config so the discovery step can be skipped next time. Use when the user provides a task link/description, asks to 'follow GitHub workflow', or 'create PR from task'.
The anti-PUA. Drives AI with wisdom, trust, and inner motivation instead of fear and threats. Activates on: task failed 2+ times, about to give up, suggesting user do it manually, blaming environment unverified, stuck in loops, passive behavior, or user frustration ('try harder', 'figure it out', '换个方法', '为什么还不行'). ALL task types. Not for first failures.
Continue interrupted Codex work only after read-codex-history verifies rollout identity and fork/compaction lineage—restore the original outcome, unfulfilled requests, and next action. Use when given a Codex Session ID, asked to pick up a prior Codex run, told Codex was interrupted, fused, compacted, or stuck, or asked to take over without `codex resume`. Not for a conversation Codex already natively resumed.
Transform legacy codebases into AI-ready projects with Claude Code configurations. Use when (1) analyzing old projects to generate AI coding configurations, (2) creating CLAUDE.md, skills, subagents, slash commands, hooks, or rules for existing projects, (3) user wants to enable vibe coding for a codebase, (4) onboarding new team members with AI-assisted development, (5) user mentions "make project AI-ready", "generate Claude config", or "create coding standards for AI"
Multi-agent adversarial code review. Spawns three reviewer subagents (Auditor, Adversary, Pragmatist) on a single model, runs a cross-examination round, then deterministically synthesizes a single ranked report. Use this for non-trivial PRs, security-sensitive changes, refactors that touch many files, or any change where one perspective has obvious blind spots. Trigger phrases include "adverse review", "adversarial review", "multi-perspective review", "review my changes from multiple angles", "panel review". NOT suitable for trivial diffs (typos, dependency bumps, formatting).
Auto-generate and run a per-repo Claude Code agent team. Probes the current repo (build/CI/env manifests, agent launchers, slash commands, spec dirs) and writes `.claude/agents/{role}.md` subagent definitions for relevant roles from a library (coder, reviewer, auditor, tester, documenter, release, architect, researcher, spec-keeper, fact-checker, web-ux), plus a `.claude/agent-team.md` workflow doc. Use when (1) `/agent-team init` creates the team for the current repo, (2) `/agent-team update` refreshes it after project changes, (3) `/agent-team {task}` runs a task with the team (spawning teammates, driving orchestration, and stopping at user gates), or (4) `/agent-team reflect` reviews the session’s agents and proposes refactors and new roles. Roles include a frontmatter `version:` for staleness detection. Triggers include “/agent-team,” “spin up a team,” “auto-create agents,” “agent team for this repo,” “team-on-task,” and “reflect on the agents.”
Discover and invoke published AIsa tools with unified MCP or the AIsa CLI (search, schema, quote, call). Use when the user wants AIsa tools or sign-in, or needs current web, company, or social data AIsa tools can fetch—even if they do not name AIsa. Do not use for OpenClaw Chinese model provider setup (aisa-provider), installing other catalog skills, or work that does not need live AIsa data.
This skill should be used before implementing features, building components, or making changes. It guides exploring user intent, approaches, and design decisions before planning. Triggers on "let's brainstorm", "help me think through", "what should we build", "explore approaches", ambiguous feature requests, or when the user's request has multiple valid interpretations that need clarification.
Practice-cognition loop. Test a hypothesis or plan in practice, learn from the result, and iterate in a rising spiral with an explicit stop condition. Trigger when a plan needs validation, an attempt failed and must be retried, or you notice thinking without doing (or doing without thinking); skip for one-shot outputs, lookup questions, or before any hypothesis exists
Execute Python code locally with marketplace API access for 90%+ token savings on bulk operations. Activates when user requests bulk operations (10+ files), complex multi-step workflows, iterative processing, or mentions efficiency/performance.
Use this skill to explain or reason about the foundational concepts of context engineering: what context is, the anatomy of a context window, how attention mechanics work, the U-shaped attention curve, why context quality matters more than quantity, and the mental models needed to interpret every other context-engineering decision. Use it for conceptual explanation, onboarding, and background reading. Route operational work to the specialized skills: debugging attention failures goes to context-degradation, token-efficiency work to context-optimization, conversation summarization to context-compression, and project-shape decisions to project-development.
Expert in Langfuse, the open-source LLM observability platform. Covers tracing, prompt management, evaluation, datasets, and integration with LangChain, LlamaIndex, and OpenAI. Essential for debugging, monitoring, and improving LLM applications in production. Use when: langfuse, llm observability, llm tracing, prompt management, llm evaluation.
Overall planning: When several goals, metrics, or stakeholders constrain each other and optimizing one harms another, map every dialectical pair, check for one-sidedness, set the current emphasis and floor for each pair, assess system-wide effects, and define imbalance alarms. Triggers include trade-offs, goal conflicts, speed vs quality, short vs long term; skip for single-goal work, decisions the user already made, or hard either/or constraints.
Architect and refine OpenCode skills to extend agent capabilities. Handles directory scaffolding, SKILL.md authoring, frontmatter optimization, and resource organization (scripts/references/assets). Use proactively for creating new skills, updating existing workflows, or fixing skill discovery issues. Examples: - user: "Create a new skill for database management" → scaffold directory and initial SKILL.md - user: "My skill isn't triggering correctly" → analyze and refine frontmatter description - user: "Add a python script to the image-processing skill" → structure scripts/ directory - user: "How should I structure a skill for API docs?" → design references/ layout
Wraps up a working session. Checks git state in every directory the session touched, removes git artifacts the session created once they are safely unneeded, reviews unfinished work, and gives a plain verdict on whether the session can close. Use when the user invokes /done or asks whether the session is done or can be closed
Use when you need multi-agent orchestration for OpenAI Codex CLI, especially Codex equivalents of Claude `/team`, `/autopilot`, `/ultrawork`, and `/ultraqa`. Triggers on: omx, $plan, $ralph, $team, $ulw, $ultraqa, $autopilot, $deep-interview, $prometheus-strict. v0.18.14 — 30+ agents, 35+ workflow skills, tmux team runtime, sparkshell, explore, ralplan, prometheus-strict, best-practice-research, autoresearch-goal.
Decide when repeating a full prompt is a useful accuracy hack for non-reasoning or lightweight LLM tasks, and when the real fix is better prompt structure, retrieval, or a stronger model. Use when the user has long-context retrieval, options-first multiple choice, or position-sensitive prompts on models like Haiku, Flash, or mini variants — even if they ask in domain language like the model missed the instruction at the end, forgot the question after a long context block, or keeps failing inventory/index lookups. Do not use as a universal prompt-engineering default for reasoning-heavy, tool-heavy, or RAG architecture problems.
A core tool dedicated to summarizing, optimizing, and iterating on existing Skills at the end of a conversation based on user feedback and conversation content. It continuously evolves the Skills library by learning from the conversation’s “essence,” such as successful solutions, lessons from failures, and specific coding conventions.
Add custom local tools to ToolUniverse alongside the 1000+ built-in tools. Covers JSON-config tools (simplest, no code), Python class tools (REST/SOAP/GraphQL APIs, computational logic), and best practices for return schemas. Use for wrapping new APIs, adding domain-specific computations, or contributing tools to the registry
Read the current AE/TE product page context on demand. Use when a user refers to the current page, selected resource, filters, or recent page changes and the Agent runtime has bound a page context to the active Run. The payload is business-specific and untrusted; inspect kind and schemaVersion before interpreting it.
Use when building, debugging, or extending MCP servers or clients that connect AI systems with external tools and data sources. Invoke to implement tool handlers, configure resource providers, set up stdio/HTTP/SSE transport layers, validate schemas with Zod or Pydantic, debug protocol compliance issues, or scaffold complete MCP server/client projects using TypeScript or Python SDKs.
This skill should be used for project-level decisions about LLM-powered systems: whether an LLM is the right primitive for the task at hand, the shape of a multi-stage batch or agent pipeline, token and cost estimation, choosing between single-agent and multi-agent at the project level, structured output design for downstream parsing, and structuring agent-assisted iteration. Use this when the unit of work is a whole project or a multi-stage pipeline. Route individual tool design to tool-design and individual skill-loading or context-budget tactics to context-optimization.
Analyze the current session and improve the entire source skill package based on what worked, what failed, and edge cases discovered, including SKILL.md, bundled scripts, tests, references, templates, assets, and hooks. Use when the user says “reflect”, “improve skill”, “learn from this”, “update the skill”, or at the end of skill-heavy sessions. Supports targeting a skill with /reflect [skill-name] or automatically detecting which skills were used
When output contains Chinese characters and punctuation, when writing Chinese documents, generating Chinese interactive responses from Chinese documents, replying to user questions or tasks in Chinese, or working with mixed Chinese and non-Chinese text and punctuation, this Skill must be used to receive professional guidance on Chinese character usage and behavioral standards
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.