Connect models and tools
Agent skills
Install ready-made skills from the skills.sh registry, add them to any chat with /, and write your own.
A skill is a packaged set of instructions for one kind of task: a research method, a writing style, a review checklist. Add it to a message and the AI follows it, so you don't retype the same long prompt in every branch.
You don't have to write them yourself. The Agent Skills page installs skills published on skills.sh, a public registry, and makes them available to Claude Code too.
Browse the registry
Open Settings and choose Agent Skills under Templates (it needs an internet connection). Skills appear in skills.sh leaderboard order:
| Column | Shows |
|---|---|
| # | Leaderboard rank |
| Skill | Name, a Verified badge for official publishers, the source GitHub repository, and a short description |
| Checks | A shield summarizing third-party security audits. Hover it for each result |
| 8W trend | Weekly installs over the last eight weeks |
| Installs | Total installs on skills.sh |
Click a row for the full description, each security check and a View on skills.sh link. Type at least two characters in Search skills to search the whole registry, beyond the leaderboard. There's no sort option.
Install, update and remove
Click Install. The skill is saved to ~/.agents/skills in your home folder, the same place the skills.sh installer uses, and linked into ~/.claude/skills so Claude Code finds it. The skill appears in the / menu right away.
To remove a skill, click Installed on its row. It's deleted immediately, without confirmation, along with its link. There's no update button: to get a newer version, remove the skill and install it again, which discards any edits you made.
Note: A chat that used a skill you later removed (and any chat that reads it through a connection) shows "Missing slash command reference" when you send, until you install the skill again.
Use a skill
Type / in a chat, open Skills, and pick one (or type part of its name). It becomes a chip in your message, and when you send, Slashspace inserts the skill's instructions at that spot. Slash commands, skills and prompts covers the rest of the menu.
Skills work with every model, including Claude Code and Codex. Agent mode never loads a skill on its own: you add it with /. Skills don't take effect in Orchestrator mode.
Slashspace sends only the text of the skill's SKILL.md file. Registry skills that rely on bundled scripts or reference files may only partly work.
Write your own skill
Create a folder in ~/.agents/skills containing a SKILL.md file that starts with a --- line:
---
name: Source comparison
description: Compare sources in a table of claims, evidence and gaps
---
When I share several sources, build a table with one row per source...name is the label in the / menu (the folder name if you leave it out), and description shows under it. Everything after the frontmatter is what the AI receives, up to 120,000 characters. Slashspace rereads the file on every send, so edits apply to your next message. Folders nested inside ~/.agents/skills work; symlinked folders and ~/.claude/skills are not read.
Note: The Agent Skills page recognizes installed skills by folder name. If your folder shares a name with a registry skill, that row shows Installed, and clicking it deletes your folder.
Review before you install
Skills are instructions written by other people, and the AI follows them. In Agent mode they can steer how the AI uses your connectors, and Claude Code can run scripts bundled with a skill. Before installing, check the security results and the Verified badge, and read the skill on skills.sh. Passing checks are a signal, not a guarantee.