Skills
96,981 packaged capabilities and playbooks for agents. Install any of them with one command.
Agent workflows
Use when executing coding tasks with a strict delivery workflow: build a full plan, implement one step at a time, run tests continuously, and commit by default after each step (`per_step`). Support explicit commit policy overrides (`final_only`, `milestone`) and optional BDD (Given/When/Then) when users ask for behavior-driven delivery or requirements are unclear.
Analyze a task, pick the right fleet type, and generate a ready-to-launch fleet (fleet.json + prompt.md files). Discovers available fleet skills dynamically. Use when the user wants to run work in parallel, asks to "plan a fleet", or says "fleet-plan".
Orchestrate depth-first mastery of ONE specific object the user already partly knows — a paper, method, algorithm, theorem, derivation, or result — until they can re-derive it, criticize it, and find where it breaks. Use when the user says "I want to really understand X", "help me understand this paper deeply", "I get the gist but not the mechanism", "walk me through this method", "defend/criticize this", or "tear this apart so I get it". Routes through singleton skills and verifies gates with graded exercises. Also supports DISCUSSION MODE, where the user masters the object by explaining it to an honest novice, an opinionated lab mate, and a professor; trigger it on "I learn better by discussing", "question me while I explain", "let's discuss instead of lecturing", "run this like a lab meeting", or "grill me on this paper". Use flow-learn-new-topic for breadth-first onboarding into an area; use flow-idea-to-proof to formalize the user's own idea.
Diagnose gateway failures by reading daemon logs, session transcripts, Redis state, and OTEL telemetry. Full Telegram path triage: daemon process → Redis channel → command queue → pi session → model API → Telegram delivery. Use when: 'gateway broken', 'telegram not working', 'why is gateway down', 'gateway not responding', 'check gateway logs', 'what happened to gateway', 'gateway diagnose', 'gateway errors', 'review gateway logs', 'fallback activated', 'gateway stuck', or any request to understand why the gateway failed. Distinct from the gateway skill (operations) — this skill is diagnostic.
Grilling session that challenges your plan against the existing domain model, sharpens terminology, and updates documentation (CONTEXT.md, ADRs) inline as decisions crystallise. Use when user wants to stress-test a plan against their project's language and documented decisions.
Repository structure methodology for maximum AI agent effectiveness. Three pillars — context engineering (repo as knowledge product), architectural constraints (deterministic enforcement), garbage collection (active entropy fighting). Use when setting up repos for AI development, diagnosing repeated agent failures, writing AGENTS.md, or designing CI gates and structural tests.
Steel-manning resolution for tempa-spec's dark mode. Resolves findings that would normally escalate to the user (DECISION/HIGH from hrt-align-consistency-review and hrt-apply-code-review; single-reviewer UNANCHORED from hrt-adversarial-authoring). Always a fresh subagent. Use only when dark.md exists and the calling skill has a finding it cannot resolve on its own evidence.
Opt-in gate for tempa-spec's dark mode (bounded autonomous decision-making across discovery-through-align, optionally through apply). Never offered proactively - only activates when the user asks for it unprompted, in their own words. Shows a mandatory risk warning before any confirmation, then asks which variant (planning-only or +implementation) and creates dark.md. Use once, at or after discovery, only on explicit user request.
Single source of truth for where an escalation-worthy finding goes when tempa-spec's dark mode is active (dark.md exists). Each of proposal, specs, design, align, and apply points here instead of inlining its own dark-mode carve-out. Says which findings route to hrt-dark-mode-decision-gate, which still go to the user, and which dark.md section records the result.
Implement and ship one agent-ready ticket unattended using its pre-agreed seams, halting instead of guessing when the ticket leaves a decision open. Use for autonomous or cloud-agent coding runs that should end in an unmerged PR ready for human verification.
Run 150+ AI apps via inference.sh CLI - image generation, video creation, LLMs, search, 3D, Twitter automation. Models: FLUX, Veo, Gemini, Grok, Claude, Seedance, OmniHuman, Tavily, Exa, OpenRouter, and many more. Use when running AI apps, generating images/videos, calling LLMs, web search, or automating Twitter. Triggers: inference.sh, infsh, ai model, run ai, serverless ai, ai api, flux, veo, claude api, image generation, video generation, openrouter, tavily, exa search, twitter api, grok
Implement comprehensive evaluation strategies for LLM applications using automated metrics, LLM-as-judge, human feedback, and benchmarking. Use when testing LLM performance, measuring AI application quality, comparing prompts/models, or establishing evaluation frameworks. Covers RAGAS for RAG pipelines, evals-as-code CI/CD integration, and modern 2025/2026 practices including structured output evaluation and agentic task success measurement.
Implement-review-fix loop for development tasks. Use when the user invokes $loop, says "loop", "loop this", "run the loop", "implement then review", "review/fix until clean", or otherwise asks Codex to keep cycling through implementation, review, fixes, and re-review until no actionable issues remain.
Top-level Lytics CDP agent - routes user intent to the appropriate skill for audience management, profile exploration, data integration, and schema browsing. Use when the user has a general Lytics question or request that doesn't clearly map to a specific skill.
A guide to verifying the Mastra framework's APIs, exploring the latest documentation and migrating versions. Use it to confirm current specifications before writing Mastra code, validate API signatures, build agent / workflow / tool / memory / RAG, and migrate such as from v0.x to v1.x. Treat internal knowledge as outdated assumptions and always verify against the type definitions of the installed version or the official documentation. For design decisions such as separation of responsibilities, use mastra-ai-architecture-rules.
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).
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).
Auto-update the show-me-the-money skill suite to the latest version. One command checks npm for a new version, downloads it, replaces the installed skills, and shows what changed. Use when the user wants to update skills, check for new versions, or says 'upgrade money', 'update skills', 'latest version', 'auto-update', or 'check for updates'.
Orchestrate two or more local workers with independent boundaries, using Orca Run/Task/Dispatch, separate worktrees/sessions, or a tmux fallback, with the PM handling decomposition, dispatch, inspection, 429 stall recovery, independent acceptance, PR wrap-up, and cleanup of temporary resources; also used when the user explicitly asks for "parallel progress", "multiple workers", "PM as overall controller", or "Wave Autopilot", or to prevent the PM from escaping into direct implementation. Do not use for a single short task, pure status syncing, or work that only needs Git branch, commit, PR, or merge rules.
Turns product or tech specs into concrete Notion tasks that Claude code can implement. Breaks down spec pages into detailed implementation plans with clear tasks, acceptance criteria, and progress tracking to guide development from requirements to completion
Shared OpenSpec workspace gate for OPSX workflows: locate project root, verify openspec/, and require exact native OPSX skills before delegation. Triggers - "OpenSpec workspace gate", "OPSX workspace locate", "OpenSpec native skill check" / OPSX workspace gate, OpenSpec project root, native skill gate. Do NOT use for Jira, Git, retry binding, or implementation orchestration.
Use the public `orca` CLI to operate Orca-managed worktrees, folder contexts, terminals, repos, automations, artifacts, worktree comments, and the browser embedded inside the Orca app. Use when the user says "$orca-cli", "use orca cli", "Orca worktree", "child worktree", "cardStatus", "spawn codex/claude in a worktree", "read/wait/send Orca terminal", "terminal send", "full handoff", "handover", "give this to another agent", "another worktree", "Orca browser", "orca artifacts", "share HTML/Markdown", "public artifact link", or "control the browser inside Orca". Prefer this over raw `git worktree`, ad hoc PTYs, Playwright, or Computer Use when the task touches Orca-managed state. Use Computer Use for browser windows, webviews, or desktop UI outside Orca's embedded browser.
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.
Multi-agent orchestration for complex tasks. Use when tasks require parallel work, multiple agents, or sophisticated coordination. Triggers include requests for features, reviews, refactoring, testing, documentation, or any work that benefits from decomposition into parallel subtasks. This skill defines how to orchestrate work using cc-mirror tasks for persistent dependency tracking and TodoWrite for real-time session visibility.
Use when you want automatic token-saving without manual skill selection. Auto-detects long sessions (>15 turns), high token usage, or bulky context — then auto-enables paleo + paleo-trim-context + paleo-converse + paleo-budget at safe defaults. User does not need to think about which skill is active. Off: "disable paleo-auto" / "manual paleo".
Use when user says "compress conversation", "condense chat", "summarize history", "too long context", or a session has many old turns. Condense prior conversation turns into a tight summary, merge near-duplicate messages, keep the last N turns verbatim. Saves context tokens without losing task state. Off: "stop condensing" / "keep full history".
Use when user says "compact json", "minify output", "tight structured", or the agent must emit JSON / structured data. Strip insignificant whitespace, collapse long arrays, shorten keys where safe, keep it parseable. Cuts tokens on structured payloads. Off: "pretty print" / "human-readable".
Use when user says "summarize output", "condense this", "tldr", or a tool result / log / file dump / long stdout is large. Reduce long content (tool output, logs, diffs, docs) to a concise intisari with key fields preserved. Saves context tokens on bulky payloads. Off: "full output" / "no summary".
Handle general Paperclip administration and small operator requests without starting a full planning session. Use when the user asks to check Paperclip state, make a minor control-plane change, update an existing Paperclip agent, inspect or attach company skills, prepare or dispatch an issue handoff, adjust assignments or reviewer gates, or perform ad hoc Paperclip maintenance.
Bootstrap skill for discovering additional skills and context from a parent project when workerd is used as a submodule. Load this skill when tasks span project boundaries (e.g., Sentry/production investigation, integration testing, cross-repo debugging).
Only use when the user explicitly invokes /plan-explain (or $plan-explain in Codex). Never auto-invoke. Reads a plan file such as those under `.sisyphus/plans/` and generates a structured summary report so a human can immediately grasp what that plan does. It only extracts facts and performs no interpretation, speculation or additional investigation.
Manage sanitized project memory and Obsidian/LLM wiki gates for large UI/UX, SEO, Figma, Stitch, progress-screen, skill-system, migration, launch, or multi-session agent projects. Use when a project needs durable context, llms-context packs, decision capture, handoff notes, or wiki sync without storing secrets.
Use this skill when the user explicitly asks to create, write, improve, or optimize a prompt for use with an AI. Trigger on phrases like "write me a prompt", "improve this prompt", "create a system prompt", "how do I ask ChatGPT/Claude to...", or "quero um prompt para...". Do NOT trigger for direct task requests where the user wants the output, not the prompt.
Generate world-class, production-ready prompts in one shot through intelligent 5-7 question flow and 69 presets across 15 professional domains. Supports XML/Claude/ChatGPT/Gemini formats with quality validation. Returns a single mega-prompt ready to use—skill does NOT implement the work described in the prompt. Optimized for modern LLMs including Claude 3.5+, GPT-4.1, and Gemini 2.5.
Create, improve and test agent skills for research workflows (paper writing, lab protocols, analysis pipelines, domain databases) that meet the Agent Skills specification and this repository's quality and security bar. Use when turning a repeated research task into a reusable skill, contributing a new skill to Research Agent Skills, rewriting an existing skill, writing trigger-accurate descriptions, or adding evals. Works alongside Anthropic's official skill-creator.
Only use when the user explicitly invokes /review-pr-check (or $review-pr-check in Codex). Never auto-invoke. A skill that orchestrates PR review handling by splitting it into a collection worker, a classification subagent, and an execution worker. Collection is delegated to the collection worker running gh pr-review-check; classification, grouping and work-packet generation to the classification subagent (the default example being Oracle); and actual work such as implementation, replies and resolves to the execution worker, while the parent manages the completeness gate, fallback triggering and the polling loop. In environments without a subagent mechanism, the parent degrades to running it itself.
Only use when the user explicitly invokes /search-history (or $search-history in Codex). Never auto-invoke. Cross-searches the local chat history (JSONL) of Claude Code / Codex CLI by keyword and presents the matching sessions and conversation fragments. History files are treated as read-only.
Use only when the user explicitly names session-handoff or explicitly asks to create a durable resume packet for a later agent session. Capture the exact live state, unresolved work, decisions, constraints, and next action needed to continue without re-deriving context; do not create a packet when there is no actionable state to transfer.
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.