✶skillsmith
a tiny workshop for non-coders

Teach your agent a new trick

A skill is a little instruction card you hand to your agent — so it knows how you like things done. We'll interview you, polish your words, and hand you back a tidy SKILL.md file.

~3 minutes · no signup
SKILL.md
brand-voice
SKILL.md
standup-notes
SKILL.md
recipe-rescuer
name: recipe-rescuerdescription: Salvage half-remembered…
## Instructions
ready
the workshop, in three movements

Talk it out. Polish it. Take it home.

01

You answer six small questions.

Name. Purpose. When to use it. Steps. Gotchas. An example. Plain English, no jargon.

02

Agent polishes your words.

One click and the muse tightens grammar, sharpens specifics, and shapes your steps into proper imperative voice.

03

Out pops a SKILL.md.

A real, valid skill file with YAML frontmatter and the right structure. Drop it in ~/.claude/skills/ and you're off.

for your agent, not just you

Or let your agent author its own skills.

Skillsmith ships as an MCP server too — the same six-step pipeline, callable from Claude, Cursor, ChatGPT, and any other tool that speaks MCP. Pick yours below, paste once, restart, done.

URLhttps://v0-hack-phi-orcin.vercel.app/api/mcp/mcp
Open in
Claude DesktopAnthropic's desktop app
1

In Claude Desktop, open Settings → Developer → Edit Config, then paste this inside the file.

{
  "mcpServers": {
    "skillsmith": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://v0-hack-phi-orcin.vercel.app/api/mcp/mcp"]
    }
  }
}
2

Save the file, fully quit Claude Desktop, then reopen it. Ask Claude to “make a SKILL.md for X” and it'll reach for Skillsmith on its own.

What your agent gains: 7 new tools
  • search_skills — look for prior-art SKILL.md files on GitHub
  • polish_skill_field — rewrite a single answer in Skillsmith's voice
  • research_skill — pull findings, pitfalls, and sources for the topic
  • synthesize_skill — produce the full SKILL.md from interview answers
  • optimize_skill_description — sharpen the description so the agent fires
  • test_skill_trigger — stress-test whether the trigger would catch real asks
  • run_skill_pipeline — do all of the above in one call

Same prompts, same model, same registry as the webpage above — agents and humans share one source of truth.

Don't see your tool? Any MCP-compatible client works — point it at the URL up top.