Skills in the command bar

If you have written skills or slash commands for Claude Code, the command bar offers them on any page. Choosing one runs it on the page in front, in that page’s chat, with the page as its input: a skill that turns an article into notes, or a command that checks a pull request, works on whatever you are reading.

Find a skill

With a tab in front, open the command bar (⌘T or ⌘L) and type the beginning of a skill’s name, with or without a leading /. Its row shows:

• the name with a slash, such as /summarize
• “Skill” or “Befehl” (command), followed by its description
• Auf diese Seite anwenden (Apply to this page) on the right

A name matches when it starts with what you type; from three letters on, also when it contains it anywhere. A space ends the match, so you cannot type words after the name here.

A / alone lists your skills and commands in alphabetical order. Typed with a /, they come first in the results, up to six; without it, up to three come after tabs and commands.

The rows only appear while a tab is in front: not over an agent chat, not with nothing open, and not in private windows.

What OpenArch reads

OpenArch looks in two folders of Claude Code’s:

Folder	What counts	Name	Description
~/.claude/skills	each folder with a SKILL.md	name in its frontmatter, else the folder’s name	description in its frontmatter
~/.claude/commands	every .md file, subfolders included	the file’s name without .md	description in its frontmatter, else its first line of text

A command with the same name as a skill is left out. Descriptions are shortened to 140 characters.

OpenArch reads only names and descriptions, never runs anything itself, and reads the folders again at most every ten seconds, so a new skill shows up without restarting OpenArch. Skills in a project’s own .claude folder and skills from plugins are not listed.

If OpenArch’s environment has CLAUDE_CONFIG_DIR, it reads that folder instead of ~/.claude.

Note

CLAUDE_CONFIG_DIR set only in your shell’s configuration (.zshrc and the like) does not reach OpenArch when you start it from the Dock or Finder. It then reads ~/.claude.

How a skill runs

Choose the row. The page’s chat opens beside the page and sends Claude Code a message that starts with / and the skill’s name, followed by:

• the page: its address, title and text, as for a question in the chat
• the selection and the tabs named with @, if the chat has any
• what is in the chat’s field, or else “Wende das auf diese Seite an.” (Apply this to this page.)

Claude Code runs it as the skill or command of that name, the way it would in the terminal, with the rest of the message as its input. The chat shows your message as /name, with chips for what went along, and the answer streams in below. Afterwards it is an ordinary page chat: ask follow-up questions, or run another skill.

A skill always gets the page, even when the chat already had it. If the chat is still answering, choosing a skill only brings up the chat; choose it again once the answer is there.

Tip

To run a skill on part of a page, select the text and press ⌘E first, so the selection becomes the chat’s focus, then choose the skill from the command bar. Words typed in the chat’s field before you choose the skill go along as its input.

What Claude gets and what it may do

Claude gets the same as for a question in the page chat: the page, the selection, the tabs you named, and your words. Claude Code reads the skill’s own file itself. The skill runs in the page chat’s folder, ~/Library/Application Support/OpenArch/PageChats, not in a project of yours.

It runs with Claude Code’s default permissions. A tool the skill would need your approval for is refused and shows in the chat with Erlauben und weiter (Allow and continue), which allows it for this chat.

How the runs work in detail: How OpenArch talks to Claude.