Skills

Pick one ready-made Claude skill and get it running first

For people who want to enable ready-made skills in Claude and related agents. The first time, use only the connect-apps plugin commands, and paste a key only after you confirm the prompt.

Official source · awesome-claude-skills

Install

claude --plugin-dir ./connect-apps-plugin

Who it is for

If you want to add ready-made task instruction packs to Claude or nearby coding agents instead of writing a spec from scratch, this collection fits. Officially it is more than a thousand reusable Claude skills and plugins for many productivity scenes. Coverage is not only Claude.ai and Claude Code; it also includes coding agents such as Codex, Cursor, Gemini CLI, and Antigravity. That is a named set, not a claim that it can only work under those names.

A skill itself is a reusable instruction pack: each skill is a folder with SKILL.md that has a YAML header (name, description), plus Markdown instructions, and it may also include scripts, references, and assets. A better fit already can open the claude command, has a working directory that contains ./connect-apps-plugin, and wants to get one official plugin path running first. Official notes:

https://github.com/composiohq/awesome-claude-skills

First try in 3 minutes

Use only the install-and-setup path written on the page. First install the plugin:

claude --plugin-dir ./connect-apps-plugin

Then run setup:

/connect-apps:setup

Prerequisites: this machine already has the claude command, and ./connect-apps-plugin really exists in the current directory. The page writes the plugin directory as a relative path and does not give a second download or market-install command. Do not guess another skill from this collection as the same installer.

Setup will ask you to paste an API key and says you can get one free at dashboard.composio.dev. Before pasting, confirm the window comes from this /connect-apps:setup. Do not write the key into the repository or the chat log. This action grants external access; paste only after the prompt looks right.

Success signal: claude accepts the plugin directory, /connect-apps:setup reaches the key step, and setup finishes after you paste, instead of saying ./connect-apps-plugin cannot be found. After the session starts, the agent should be able to see each skill's name and description.

Usage examples

Example 1: install the connect-apps plugin and finish setup first.

Officially the steps are two parts: install the plugin with claude --plugin-dir ./connect-apps-plugin, then run /connect-apps:setup. The first time, stop when setup ends and the key is accepted. The page does not write the next business command after a successful setup, and it does not write which apps this plugin will connect. What you can confirm is this install-then-setup order, and that the key comes from dashboard.composio.dev.

Example 2: load one skill folder on demand instead of reading the full text at once.

Officially skills load progressively. When a session starts, the agent only sees each skill's name and description, about one hundred tokens per skill. The full SKILL.md body is usually under five thousand tokens and loads only when the agent judges it relevant to the current task. The first time, you can only watch: after a related task appears, does the agent start following the instructions in that SKILL.md. The page does not name one of the thousand-plus skills as a required example, and it has no second install command beside connect-apps.

Do not invent a market install in parallel on either path. Other entries in the collection are only written as reusable instruction packs; triggering still depends on name and description.

Limits

Copyable official commands only cover the connect-apps plugin directory and /connect-apps:setup. The relative path means you must already be in a location that contains that directory. The page does not write how to obtain the directory.

Setup will ask for an API key. Officially you only paste the key from the prompt and are pointed to a free key at dashboard.composio.dev. It does not write other key placeholders, and you must not commit the key to the repository.

At session start, skills do not load the full text; they only expose name and description. If the task description does not match description, the full instructions may never appear. The page does not write an install command for each of the thousand-plus items, and it does not write a separate trigger for every agent outside Claude.ai and Claude Code.