Generate one or more standalone Meta image-ad creatives via ChatGPT Image 2 (gpt-image-2) through the Arcads external API. Locks the model, auto-strips platform chrome, enforces edge-safe layouts and glyph-safety inside body text. Use when the user asks for a "gpt-image-2 ad", "ChatGPT Image ad", "Image 2 ad creative", "make a static image ad with GPT", or anchors on a need for typography-heavy / dense-text / UI-mimicry ad creatives (chat threads, comparison tables, fake search results, iOS dialogs, Slack snapshots, ChatGPT-conversation ads, Apple Notes lists). Does NOT trigger on Nano Banana cues — use nano-banana-image-ad for those.

Clone an existing video ad for a different product or offer. Analyzes the source video's style, pacing, camera work, dialogue, and tone, then adapts and generates a new Seedance 2.0 video customized for the user's product. End-to-end workflow: input video → analysis → adapted prompt → generation → delivery. Use when someone says "clone this ad", "make this ad but for my product", "recreate this video for my brand", or provides a video ad and a product image asking for a similar video.

Creates and retrieves AI video and image-related assets via the Arcads external API (Seedance 2.0, Sora 2, Veo 3.1, Kling, Grok Video, Nano Banana, b-roll, scene, script/actor flows). Loads prompts from the bundled prompting guide and model library, respects HTTP Basic auth from ARCADS_API_KEY, and polls assets/videos until ready. Use when the user mentions Arcads, external-api.arcads.ai, Seedance, Sora2, Veo, Kling, Nano Banana, b-roll, UGC scripts, or generating marketing creative through Arcads.

Publish finished creatives as live Meta (Facebook/Instagram) ads via the Meta Marketing API, plus research and ad-copy support. Uploads an image or video, builds a multi-variant TEXT_LIQUIDITY creative, and creates a PAUSED ad in an existing ad set. Also pulls top-performing ads (ranked by ROAS) and competitor ads from the Ad Library to inform copy. Use when the user asks to deploy / publish / launch a creative as a Meta or Facebook ad, build a Meta ad, push a video or image into an ad set, pull their top ads, or research competitor ads. Not for generating creative (use the image/video skills) and not for writing AdTable/Airtable rows (use adtable-light).

Interactive quiz tutor for Obsidian StudyVault learning. Use when the user wants to: (1) Take a diagnostic assessment of their knowledge, (2) Study or review specific sections/topics, (3) Drill weak areas identified in previous sessions, (4) Check their learning progress or dashboard, or says things like "quiz me", "test me", "let's study", "/tutor", "학습", "퀴즈", "평가".

Transforms knowledge sources into an Obsidian StudyVault. Two modes: (1) Document Mode — PDF/text/web sources → study notes with practice questions. (2) Codebase Mode — source code project → onboarding vault for new developers. Mode is auto-detected based on project markers in CWD.

Create animated flow diagrams from articles, workflow notes, architecture sketches, or process descriptions using a JSON specification and a local Python/Pillow renderer. Use when the user wants to turn source material into a static PNG + animated GIF diagram.

Set a repo up to prove engineering-task work actually works before it ships. Investigates the repo, ensures a one-command dev stack (`dev-local`) exists, asks whether verification runs locally or in a sandbox (crabbox), confirms/installs the driver (the `playwright-cli` skill for web by default). Outputs three artifacts: a committed `/verify` skill (per-task verification SOP — spawn a verifier sub-agent → drive the app → screenshot/video proof → open a PR with the proof embedded), the `/dev-local` skill + script, and the installed driver skill. Use when someone says "set up verification", "make this repo verifiable", "scaffold a verify skill", "set up the verifier".

Master skill — set up the full agent harness for any repo so an agent can work it reliably: legible (map-not-manual docs + custom lints), executable (one-command dev stack), verifiable (e2e gate + a verify-before-ship loop), plus commit hygiene and entropy control. Use when onboarding a new/unfamiliar codebase to agent-driven development — "set up the harness", "make this repo agent-ready", "harness this codebase".

Use when deciding WHERE to point SEO effort, not how to write a page. Triggers: a new site or brand with no rankings and no authority ("cold start", "starting from zero", "nobody knows us"), deciding what to double down on, a page or cluster that ranks but earns nothing, hunting emerging or newly-coined keywords before competitors arrive, "should we build a cluster or one page", "what do we write next", "we get impressions but no clicks", "our traffic plateaued despite publishing", whether to chase a head term at all, or turning any of it into recurring automation. Also use when asked why an SEO effort stalled despite consistent output. NOT for writing an article, keyword expansion mechanics, or auditing a single tactic for penalty risk.

Delegate tasks to ANY CLI agent (claude, codex, aider, ...) running in a detached tmux session, with a race-safe done-signal protocol and multi-turn iteration. Use when delegating work to a non-Claude CLI agent, when the user says "tmux delegate", "run agent in tmux", "delegate to codex/aider", or when executor work should run in an observable background terminal instead of the Agent tool.

Spin up a new loop (domain) in a file-based knowledge base — bootstrap the substrate if it's missing, gather the loop's charter, scaffold domains/<loop>/README.md, then do ONE real test run and record it in the loop's Timeline and LOG.md. Use when the user says "set up a new loop", "create a domain", "start a new beat/workstream", or names a recurring job they want the agent to own.

Set up an end-to-end test suite in any repo, following practices that make e2e a reliable per-PR gate: real flows over bypass, layered assertions, a reusable auth/session helper, video+trace evidence, and a compounding suite. Use when a repo has no e2e (or weak e2e) and you want system-level tests — "set up e2e", "add end-to-end tests", "scaffold a test gate".

Scaffold a one-command `dev-local` launcher for ANY codebase. Investigates the repo to find its services, ports, and infra dependencies, then generates a single `scripts/dev-local.sh` (up/down/status/logs/restart) that runs every dev server in one tmux session, plus a short skill doc describing it. Use when someone says "set up dev-local", "make a one-command dev launcher", "I want one script to start this repo", "scaffold dev-local for this project".

Scaffold an isolated CLOUD dev box per agent (via crabbox + Daytona) for any codebase — the parallel-safe counterpart to dev-local-setup. Each agent gets its own full stack (own DB + dev server) and an in-box browser for e2e, so concurrent loops never collide on ports/state. Sets up the snapshot image, .crabbox.yaml, an idempotent setup.sh (also boots the stack locally), and a cbx.sh wrapper. Use when the user says "set up crabbox", "give each agent its own box", "add cloud testing", "make this repo testable in the cloud / on Daytona", "parallel-test this", or when setup-codebase-harness needs true per-agent isolation.

Audit a repo's agent context — CLAUDE.md files, codebase docs, skills, and tool/MCP designs — against Anthropic's Claude 5 context-engineering guidance ("unhobbling": Anthropic cut ~80% of Claude Code's system prompt with no eval loss). Finds overconstraint, conflicting instructions, redundancy, stale facts, and missing "unknown knowns"; produces a scored findings report with concrete rewrites, then applies approved fixes. Use when someone says "audit my CLAUDE.md", "context audit", "unhobble this repo", "review our agent docs/skills/tools", or after upgrading to Claude 5-generation models.

Guide for creating effective skills. This skill should be used when users want to create a new skill (or update an existing skill) that extends Claude's capabilities with specialized knowledge, workflows, or tool integrations.

IMMEDIATELY USE THIS SKILL before answering any task - internal design thinking that identifies the question type, selects applicable skills, and converges on the best approach before implementation

Use when user asks to "read paper", "analyze paper", "summarize paper", "读论文", "分析文献", "帮我看一下这篇paper", "论文笔记", or provides a PDF file that appears to be an academic paper. Specialized for CV/DL papers. Also supports Zotero integration: "读一下这篇论文 ...", "快速看一下这篇论文 ...", "批判性分析这篇论文 ...", "读一下 Zotero 里的 XXX", "批量读一下 Zotero 里 VLA 分类下的论文" **重要触发词**: "读一下 XXX"、"读一下这篇"、"帮我读" → 必须调用此 skill

重新生成 Obsidian 里的目录页 / 导航页（MOC）。 当用户说“更新索引”“更新论文和概念目录”“刷新论文和概念目录”“刷新MOC”时使用。