Skills
97,501 packaged capabilities and playbooks for agents. Install any of them with one command.
Agent workflows
End-to-end Jira bug-fix workflow (stages 0-10), driven by a single Jira link, from intake through PR/MR merge and Jira writeback. Manual mode (default) pauses for confirmation between stages; auto/force modes run end-to-end; ai-proxy overlay (thin freeze then occupy) is available independently. Triggers — 「修复这个 bug [URL]」「帮我修复 [URL]」「jira-fix [URL]」「自动修复 [URL]」「强制修复 [URL]」「继续修复」「从上次继续」；「ai-proxy 模式」「AI 代理模式」「切换 ai-proxy」 / fix this bug, jira-fix, auto fix, force fix, resume fix, ai-proxy mode, switch to ai-proxy. Do NOT use for batch fixes across multiple issues — use jira-fix-queue instead (it enqueues into goal-driven-queue; confirm, then run the queue).
Provides MoAI-ADK foundational principles including TRUST 5 quality framework, SPEC-First DDD methodology, delegation patterns, progressive disclosure, agent catalog reference, and token budget management (absorbed from moai-foundation-context). Use when referencing TRUST 5 gates, SPEC workflow, or context window optimization.
Constraint transparency: analyzes an artifact structurally, then analyzes what its own analysis concealed. Produces a conservation law AND a constraint report showing what was maximized, what was sacrificed, and what to investigate next. The only AI skill that knows what it can't see.
Authoritative reference for Anthropic products. Use when users ask about product capabilities, access, installation, pricing, limits, or features. Provides source-backed answers to prevent hallucinations about Claude.ai, Claude Code, and Claude API.
Run Rejudge — send a question to separate reviewers and a judge, returning one reviewed result (the rejudge tool inside Pi, else the rejudge CLI; read-only). Use when the user says /rejudge or wants a multi-model review. For one external reviewer, use ask-agent-cli instead.
Set up or audit a machine's coding-agent harnesses from one rule table and the user's workstation repo: the shared global instructions every harness reads; the global rules each enforces as deny, ask, or allow-and-report; the pre-tool hook; memory kept off; each harness's declared config such as its default model; the plugins each harness gets; and the skills and tools the workstation repo lists to install. Use for a new machine or VPS, to audit this machine's agent setup, or when another skill says to check the machine.
Take one feature request from prompt to reviewed, audited delivery in a SINGLE autonomous pass — one runnable Workflow that plans, reviews the plan, builds task by task, cross-reviews the build, and optionally runs a launch audit, then RETURNS the verified findings rather than looping (the user decides any fix round; a Workflow can't ask mid-run, and an autonomous fix-loop is what caused multi-hour grinds). It chains the existing skills — plan-to-task-list-with-dag (a specialist agent per task) → plan-founder-review → a specialist engineer/reviewer build across the DAG → a full claude ∥ codex/kiro cross-review → go-live-audit — and FOLLOWS the graph strictly: every task declares its dependencies, layers must be a topological order, plan review blocks a mis-ordered graph, and a task NEVER builds before its dependencies integrate (no building on a broken base). Per-task review is SLICE-SCOPED (judges only that task's own change against its acceptance criteria, attributing whole-codebase end-state gaps to the task that owns them) so the build isn't drowned in false blocks, and impl review is the end-state gate. Gates FAIL CLOSED — a reviewer that didn't actually run is never counted clean, and a died codex/kiro reviewer falls back to a native review, never a fabricated block. Durable and resumable: each run writes a live status file (overall + per-phase + per-task) with a journal reader for status/stop/resume, and resume rebuilds integration state from GIT ground truth so an interrupted run never loses or re-loses already-landed tasks. Code writing and each review independently pick native / codex / kiro (reviews can skip; writer and reviewer can differ), so the user controls quality vs token cost.
[user/auto] Close user-owned decisions one at a time with evidence-first questions. Use it for an explicit invocation, an exact active tk-drive decision handoff, or an answer to this skill's pending question. Do not auto-start on generic ambiguity, artifact existence, or generic continuation alone.
Standardized AI-assisted product delivery workflow for turning a new idea or messy repository into a disciplined build loop with `design-document.md`, `tech-stack.md`, `implementation-plan.md`, and a persistent `memory-bank/`. Use in Codex or Claude Code when bootstrapping greenfield work, normalizing a repository into a document-first process, or executing implementation one validated step at a time with `progress.md` and `architecture.md` kept current.
Use when authoring, reviewing, or debugging harness-backed vitest-evals suites, custom Harness adapters, first-party ai-sdk or pi-ai harness integrations, judges, replay, reporter-facing normalized run data, or examples and docs for these APIs.
Use when the user wants to add an in-site feedback loop to a website repo — a floating "提个建议" button where allowlisted visitors submit suggestions that become a GitHub Issue, which GitHub Actions turns into an automatic code change via Claude Code, auto-merges and deploys, and records on a /_feedback dashboard with one-click revert. Triggers — "给网站加个反馈对话框", "提一句话就自动改网站", "装上反馈闭环", "feedback loop", "/wjs-looping-feedback".
Interactive teaching agent for the goal-seeking agent generator and eval system. Provides a structured 14-lesson curriculum covering agent generation, SDK selection, multi-agent architecture, progressive evaluation (L1-L12), retrieval strategies, intent classification, math code generation, self-improvement loops with patch proposer and reviewer voting, and memory export/import.
Enter explore mode to investigate ideas, systems, problems, and requirements before planning. Use ultra mode, explicitly or through config, when the user wants a durable adaptive research bundle with an index and only the justified C4, ADR, or dependency artifacts.
Turns a sequential, ready-for-dev story backlog into conflict-free CONCURRENT WAVES. Builds a dependency DAG from epic order, per-story dependency maps, and Owned File/Module Scope overlaps, then topologically sorts it into parallel waves of mutually disjoint, dependency-satisfied stories (capped by maxParallel), and emits a parallelization-plan.md with per-story git-worktree branch names and an ordered merge sequence. Use when the user says "plan parallel work", "which stories can run in parallel", "parallelize the backlog", "build the wave plan", "parallelization plan", "conflict-free workstreams", "what can we run concurrently", "split into worktrees", "dependency graph for the stories", "merge order", or "how do I fan this backlog out to multiple dev agents". Run AFTER stories are ready-for-dev (scrum-master) and an architecture exists. This skill PLANS parallelism only — it does NOT run agents, spawn worktrees, write code, run tests, or perform git operations.
Use when building MCP (Model Context Protocol) servers to enable LLMs to interact with external services through well-designed tools. Triggers include requests to create MCP servers, integrate external APIs or services, or build MCP tools in Python (FastMCP) or Node/TypeScript (MCP SDK).
This skill should be used when the user asks to "create a plugin", "make a GitHub plugin", "turn my repo into a plugin", "create a Claude Code plugin", "create a Codex plugin", "publish my skills as a plugin", or "set up a plugin marketplace". Scaffolds the complete file structure for a repo that works as both a Claude Code and Codex plugin with marketplace support.
Self-improving knowledge wiki with a vault of raw sources. Use when the user mentions 'curiosity engine', 'wiki', 'vault', 'knowledge base', 'ingest', 'iterate', 'refine', 'improve', 'evolve', 'curator', 'lint', or wants to add sources, query accumulated knowledge, check wiki health, or run autonomous improvement. Also triggers on 'add to vault', 'what do I know about', 'improve wiki', 'set up knowledge base', 'new knowledge base', 'run curator'. Use even without explicit naming — if the user wants to file something for later or asks about accumulated knowledge, this is the skill.
Keeping a design system in an LLM's context: loading tokens compactly, persisting design decisions across sessions, handling multiple variants, and budgeting the context window. Use when generated UI drifts from the design system, or when setting up tokens and decisions so every session and agent uses the same ones.
Take real control of the desktop - list and focus windows, type, paste, click, scroll, and screenshot - on Windows, macOS or Linux, and drive other coding-agent sessions running in terminals. Use when asked to set up the screen or the day, open and arrange a set of apps or repos, prepare or run a live demo before recording, test a desktop application that has no headless harness, launch or steer a Claude Code session in another window, or capture what is on screen as evidence. Triggers on "set up my screen", "get my demo ready", "drive the screen", "control my desktop", "open these repos and start", "test this desktop app", "run this on my machine and show me". Not for browser automation, which has its own headless tooling.
Build event-driven pipelines that connect a zero-LLM mechanical layer to a duoduo brain job via session notify. Use when the user wants to monitor an external data source (feeds, prices, files, webhooks) and only wake an agent when something worth acting on actually arrives — keeping LLM costs at zero during idle periods. Also trigger for: 机械采集, 事件驱动, 按需唤醒, 数据监控, keepalive job, 外部触发, 只在有料时干活, 省 token.
CodeStable is an AI coding workflow for serious software engineering that keeps humans in the loop. Rather than centering on Agent orchestration, it organizes requirements, architecture, features, issues, and historical decisions to make development driven by Codex and Claude controllable, traceable, and continuously evolvable.
CodeStable is an AI coding workflow for serious software engineering that keeps humans in the loop. Rather than centering on Agent orchestration, it organizes requirements, architecture, features, issues, and historical decisions to make development driven by Codex and Claude controllable, traceable, and continuously evolvable.
Search, install, update, and remove third-party Agent Skills from skills.sh or any GitHub/GitLab repo via the `npx skills` CLI. Use when the user wants to "find a skill", "install a skill from github", "search skills.sh", or add third-party capabilities to Claude Code/Cursor/other agents. Not for authoring new skills (see create-skill).
Generates or updates a resumable cold-start handoff package from an in-progress conversation, review, debugging session, or investigation, with transcript materialization, verified structured artifacts, update-mode backups, and bounded review repair. Use when the user says create a handoff doc, save this for later, document what we found, update the resumption file, or prepare a fresh agent to resume without chat history.
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.