Quickstart for AI agents

This page is written for AI coding agents such as Claude Code, Codex, Cursor and Gemini CLI. If you are a person, give this link to your assistant and ask it to set up SkillGild for you.

#Your goal

Connect the user's machine to SkillGild, then use hosted skills to complete their tasks. Skills run on SkillGild's servers. You send a JSON input and receive a text result, which you apply with your normal tools.

#Rules

1. Never ask the user to paste a key, password or token into the chat. Sign-in happens in their browser. The credential is saved in the operating system's credential store.
2. Never write an API key into an MCP config, environment file or repository.
3. Ask before running each setup command, and explain it in one plain sentence.
4. Send only what the skill needs. Do not include secrets, credentials or unrelated private files in a skill's input. Tell the user that the input goes to SkillGild and its model provider if they have not already agreed to that.
5. Show changes before applying them, and ask before anything destructive.

#Set up

#1. Check for an existing connection

skillgild status

If this prints Connected to SkillGild as …, skip to step 4. If the command is not found, install the CLI first.

#2. Install the CLI

Install the skillgild CLI by following Install the CLI. Then find its absolute path, which you need in step 4:

command -v skillgild        # macOS and Linux
where.exe skillgild         # Windows

#3. Sign in

skillgild login

The command prints a URL and a code such as ABCD-EFGH, and opens the browser. Tell the user:

Please check that the code in your browser is ABCD-EFGH, sign in if asked, and click Approve.

Keep the command running. It finishes on its own with Connected. after the user approves. Codes expire after 10 minutes; if that happens, run skillgild login again. Confirm with skillgild status.

#4. Register the MCP server with yourself

Use the absolute path from step 2 with the command for your own client:

Most clients load new MCP servers when a session starts. Ask the user to restart or reload your session, then check that the tools skillgild_search_skills and skillgild_run_skill are available.

#Use skills

#Discover

Call skillgild_search_skills with a short query that describes the task:

{ "name": "skillgild_search_skills", "arguments": { "query": "accessibility" } }

Each result includes slug, name, description, access_tier, free_runs_per_month and input_schema. Pick the skill whose description matches the task. If none fits, do the task yourself.

#Install (optional)

Installing writes a small SKILL.md that helps you recognize the skill in future sessions. It contains no private implementation.

skillgild install accessibility-audit --agent claude-code

Use --agent codex, cursor or gemini-cli to match your client, or shared for a shared location.

#Run

Build an input object that matches the skill's input_schema, with every required field and the declared types. With an open schema, send {"prompt": "…"} containing a complete, self-contained task description.

{
  "name": "skillgild_run_skill",
  "arguments": {
    "skill_id": "accessibility-audit",
    "input": { "prompt": "Audit this signup form for WCAG 2.2 issues: <relevant markup>" }
  }
}

The result contains output, the skill's answer as text, and usage, the runs used against the monthly limit and the reset time. Apply output with your file tools and show the user a diff.

#Run from a shell

If MCP is unavailable, use the CLI. Pass a new lowercase UUID as the idempotency key, and reuse the same key only when retrying the same request.

skillgild run accessibility-audit --input '{"prompt":"…"}' --idempotency-key 8d9ba436-a430-44af-9d12-f1d3d4198227

#Handle errors

#Machine-readable entry points