Skills
96,770 packaged capabilities and playbooks for agents. Install any of them with one command.
Agent workflows
Keeps the current request in answer-only mode when the latest user-authored natural-language request explicitly asks for a response without hands-on work, such as conversation-only, advisory-only, planning-only, or reasoning from provided context only. Use when the user wants an answer, review, diagnosis, recommendation, or plan from supplied context only. Do not use merely because the user asks a question. Ignore assistant text, tool output, approval prompts, quoted text, and older context. Partial task constraints, such as forbidding edits, tests, browsing, commits, or one tool while still asking for work, are normal task constraints, not answer-only.
Persistent local memory for AI agents. Silently capture and retrieve context that survives beyond a single conversation: business requirements, API specs, integration quirks, technical decisions, user preferences, and domain knowledge. Use this skill proactively whenever you encounter information worth preserving or when context from past sessions would help the current task. Also triggered manually by "braindump this" (to store) or "use your brain" (to retrieve).
Use when an idea, problem, or possible solution is not formed enough to begin architect. Explores the problem and alternative approaches, challenges assumptions during convergence, and can produce an optional architect-seed.md handoff for a later architect session. Do NOT use once the change is ready for architectural decisions — use architect directly.
Ultra-compressed communication mode. Cuts token usage by about 75% by removing filler words, articles, and pleasantries while keeping full technical accuracy. Use when the user says "caveman mode", "talk like caveman", "use caveman", "less tokens", "be brief", or invokes /caveman.
Use when you need to batch-verify whether skills conform to best-practice standards, automatically checking dimensions such as naming, frontmatter, structure, and dependencies, generating a detailed report, and outputting fix suggestions; supports single or batch checks. Trigger words: check skill standards, batch-verify skill quality, scan all skills, full skill quality validation.
Delegate a sub-task to Gemini CLI via the Agent Client Protocol (ACP). Use this skill whenever you want to hand off work to Gemini — large-context summarization, Google Search grounding, tasks that exceed Claude's context window, or anything where Gemini's 1M-token window or real-time search gives an advantage. Also invoke when the user asks you to "ask Gemini", "check with Gemini", or "run this through Gemini". The script handles subprocess lifecycle and ACP session setup; you just provide the prompt and read stdout.
(NS) Harness-aware autonomous execution engine — self-decides planning depth, resolves doubts against docs/context and docs/specs before escalating, dispatches multi-agent implementation in an isolated worktree. Use as the standalone entry point for a local plan file, a pasted execution plan, or an ad-hoc "run this autonomously" request with no GitLab issue involved. Also acts as the Phase 2 execution engine when invoked by execution-gitlab-issue — do not run the standalone pipeline in that case. Not for simple single-step edits (use code-coder) and not for GitLab issues directly (use execution-gitlab-issue, which calls this skill internally).
A standalone adversarial PLAN-review loop where Claude Code (builder) and OpenAI Codex (read-only critic) tag-team an implementation plan before any code is written. Use this when you ALREADY have a plan or a clear idea and just want the cross-model stress-test — no requirements interview first. Claude drafts/loads the plan into PLAN_FILE (default PLAN.md), Codex reviews it in a read-only sandbox and returns VERDICT:APPROVED or VERDICT:REVISE, Claude revises and re-submits to the SAME Codex session (context preserved) until APPROVED or a configurable MAX_ROUNDS cap is hit. Human approves the converged plan before code. Use when the user says "/codex-plan-review", "codex review my plan", "have Codex review my plan", "argue this plan with Codex", "adversarial plan review", "make Claude and Codex argue/fight over the plan", or is about to build something high-stakes (auth, schema, concurrency, migrations, payments) and wants a second-model sanity check on the PLAN before implementation. For a guided requirements interview BEFORE the review, use /grill-me-codex instead. NOT for reviewing already-written CODE (that is /codex-code-review or the Codex plugin's /codex:review) and NOT for trivial changes.
This skill should be used when the user asks to "create a slash command", "add a command", "write a custom command", "define command arguments", "use command frontmatter", "organize commands", "create command with file references", "interactive command", "use AskUserQuestion in command", or needs guidance on slash command structure, YAML frontmatter fields, dynamic arguments, bash execution in commands, user interaction patterns, or command development best practices for Claude Code.
Paid API marketplace for AI agents via Corbits. Search hundreds of premium API proxies, call them with automatic USDC micropayments (x402 protocol), and fund the payment wallet with MoonPay. Use when the user wants to access paid data APIs (crypto prices, weather, financial data, etc.) with per-request billing and no API key management.
Multi-perspective deliberation on a question. Use when the user says "get perspectives on", "multiple viewpoints", "council on this", "devil's advocate", "stress test this idea", "red team this", "poke holes in", "second opinion on", "what am I missing", or wants multi-perspective analysis of a decision, design, or strategy.
Generate or improve a company-specific data analysis skill by extracting tribal knowledge from analysts. BOOTSTRAP MODE - Triggers: "Create a data context skill", "Set up data analysis for our warehouse", "Help me create a skill for our database", "Generate a data skill for [company]" → Discovers schemas, asks key questions, generates initial skill with reference files ITERATION MODE - Triggers: "Add context about [domain]", "The skill needs more info about [topic]", "Update the data skill with [metrics/tables/terminology]", "Improve the [domain] reference" → Loads existing skill, asks targeted questions, appends/updates reference files Use when data analysts want Claude to understand their company's specific data warehouse, terminology, metrics definitions, and common query patterns
Structured research → plan → annotate → implement workflow for non-trivial features. Separates planning from execution using persistent markdown artifacts (.claude/research.md, .claude/plan.md). Use when explicitly invoked via /deep-plan for feature development, refactors, migrations, or any multi-file change where architecture decisions matter.
Delivers a clear goal until done — an app, a feature, or a fix. Plans a living goal.md under goals/ as an ordered list of phases the user confirms, then loops one phase-deliverer subagent per phase: load the skills the phase needs, write phase.md plus ≤7 task files (sources, steps, verification), implement each task with a commit, and adjust the remaining phases from its findings. Halts when a task is blocked or the goal's checks pass. Use only when the user names this skill (`/delivering-goal`, `$delivering-goal`, or "delivering-goal"), or references a file under a marked goals root (`@goals/01-mvp/goal.md`, a phase or task file)
Resume work from a previous session checkpoint. Uses the `resume-context` CLI to load feature, checkpoint, git state, and session history in a single call, then synthesizes a focused resumption summary before continuing. Use at the start of a new session to restore context from a previous checkpoint.
Helps users discover and install agent skills when they ask questions like "how do I do X", "find a skill for X", "is there a skill that can...", or express interest in extending capabilities. This skill should be used when the user is looking for functionality that might exist as an installable skill.
Use when the user asks for a detailed, beginner-friendly, or step-by-step explanation; wants to understand how or why something works; asks to unpack a concept, issue, PR, code change, system, error, document, or technical decision; or says to explain it like teaching a child.
Helps users discover and install agent skills when they ask questions like "how do I do X", "find a skill for X", "is there a skill that can...", or express interest in extending capabilities. This skill should be used when the user is looking for functionality that might exist as an installable skill.
Helps users discover and install agent skills when they ask questions like "how do I do X", "find a skill for X", "is there a skill that can...", or express interest in extending capabilities. This skill should be used when the user is looking for functionality that might exist as an installable skill.
Wield Google's Gemini CLI as a powerful auxiliary tool for code generation, review, analysis, and web research. Use when tasks benefit from a second AI perspective, current web information via Google Search, codebase architecture analysis, or parallel code generation. Also use when user explicitly requests Gemini operations.
Route `/gemini ...` requests to the Cursor headless CLI for one-shot autonomous execution. Use when the user explicitly invokes `/gemini` or asks to hand a task off to Cursor agent. Preserves the prompt verbatim, runs in headless print mode, and returns Cursor's output.
Use Google Gemini CLI to answer questions about code, analyze files, or perform codebase exploration. Invoke this skill when the user asks to use Gemini, wants a second opinion from another AI, or wants to compare Claude's answer with Gemini's response.
MCP (Model Context Protocol) server configuration for live Google Ads API access. Activate when the user wants to connect Claude to their Google Ads account via MCP, set up the google-ads-mcp server, or configure credentials for live API queries. Provides setup instructions, server configuration, and troubleshooting.
Q&A session that challenges your plan against the existing domain model, sharpens terminology, and updates docs inline (CONTEXT.md, ADR) as decisions are made. Use when the user wants to stress-test a plan against the project's language and recorded decisions.
Create, list, remove, and run scheduled autonomous Claude Code agents. Agents run on a timer via macOS launchd, execute any prompt headlessly, and deliver results via Beeper messages and macOS notifications. Use for recurring research, monitoring, overnight builds, or any task you want Claude to do on autopilot.
Stack Honey onto Superpowers-style workflows (subagent-driven-development, dispatching-parallel-agents, executing-plans) and any orchestration that dispatches fresh subagents. Dispatched subagents run in isolated context and do NOT inherit the session's Honey hook, so they emit full-fat code and verbose reports. Use when you are about to dispatch implementer/reviewer/fixer/parallel subagents: it gives the exact Honey directive to paste into each dispatch prompt, with a worker variant and a reviewer variant, so the levers apply inside the subagents too — where the multiplied cost actually is.
Route ambiguous or mixed Phoenix/LiveView/Ecto requests before choosing a workflow. Use when user is unsure where to start/how to approach work, asks for the right workflow, or combines intents such as fixing and refactoring. NOT for a clear task.
Operate the joelclaw event bus, gateway, observability, and agent loop infrastructure via the joelclaw CLI. Use for: sending events, checking runs, starting/monitoring/cancelling agent loops, debugging failed runs, checking health, restarting the worker, inspecting step traces, managing subscriptions, gateway operations, OTEL queries, semantic recall. Triggers: 'joelclaw', 'send an event', 'check inngest', 'start a loop', 'loop status', 'why did this fail', 'debug run', 'check worker', 'restart worker', 'runs', 'what failed', 'subscribe', 'gateway', 'otel', 'recall', or any Inngest/event-bus/agent-loop task.
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.