Skills

Agent skills I use across my own products (AdaptlyPost, Flowsery, RedReplier, ClawOneClick). Each one is a SKILL.md folder, so it works with Claude Code, Codex, Cursor and any agent that reads skills.

Every skill here came out of shipping the thing first and writing down what actually worked. No theory, no guesses about what Google or a framework "probably" does.

Pages for each skill, with a copy button, live at tarasshynkarenko.com/skills.

Install

npx skills@latest add TarasShyn/skills

Or clone one folder by hand into .claude/skills/<name> (or your agent's skills directory).

Reference

• google-preferred-sources-button — Add a "Make us preferred on Google" button to any site so readers can mark it as a Google preferred source. Placement, wording, framework-agnostic code, tests. Everything factual comes from Google's own guide.
• mcp-directory-submission — Publish a remote MCP server everywhere agents look: the official MCP registry, Smithery, Glama, the Claude Connectors Directory and plugin marketplace, ChatGPT apps, both punkpeye awesome lists, Cursor, xAI, and the paid mcp.so. The public repo layout every directory reads, the OAuth and CORS requirements, scripts, and the rejections from three production servers.
• entity-stack-kit — Read a site's sameAs schema and generate one HTML submission kit: logo, banner, descriptions, and every entity profile with copy buttons and per-directory guides. Machine-readable for agents; optional DataForSEO or Ahrefs description alignment.
• agent-markdown — Serve a Markdown twin of every indexable page to AI agents: .md suffix URLs, Accept: text/markdown negotiation, /sitemap.md, llms.txt, and 39 production gotchas from shipping it on Next.js + Cloudflare Workers. Moved here from TarasShyn/agent-markdown.
• worktree-manager — Route every git worktree an agent makes through one script: .worktrees/<name> on its own branch, node_modules linked, .env files copied, a port slot per worktree, and a remove that won't throw away unmerged work. Installs Claude Code's WorktreeCreate/WorktreeRemove hooks and a guard that blocks raw git worktree add.
• stfu — Ask a short question, get the answer with zero tool calls. Stops the agent from re-running grep, git show or ssh to "double check" work it did two minutes ago. Built from 30+ "ANSWER THE QUESTION" interrupts in a month of my own sessions, most of them after a single grep.

License

MIT