SkillCatalog for AI agents

AI agents such as Claude Code and Codex can run skc for you. This page tells an agent what to read about SkillCatalog and how to run skc so that it can read the results and nothing changes without your approval. Point your agent to this page.

What to read

Source	What it answers
The Essentials skills, starting with `sc-overview`	What SkillCatalog is for, and which skill to read next
`skc help-json`	Every command and flag of the installed `skc`, as JSON
https://skillcatalog.dev/llms.txt	An index of these docs. https://skillcatalog.dev/llms-full.txt holds every page in one file
https://skillcatalog.dev/schemas/index.json	The JSON schema of each file format you write, such as `SKILL.md` front matter and the team manifest

If the user added the Essentials skills during the desktop app's first-launch setup, you already have them. The website describes the latest release, so when it disagrees with the installed skc, trust skc.

How to run skc

Pass --json. Each command then prints one JSON object: to standard output when it succeeds, and to standard error when it fails or finds a problem. Read its outcome, not the exit code, because the same exit code means different things in different commands.
Pass --catalog-id to scoring commands. Without it, they use the catalog last chosen in the desktop app.
Pass --message to skc sync. Without a terminal, sync refuses to commit unless it has a message.
Run project commands at the top of the checkout. skc install, skc update, and skc uninstall work only on the folder where they run and never search parent folders.
Ask the user first before a command that commits, pushes, delivers, or deletes, such as skc sync, skc deliver, skc profile add, or skc catalog remove. Checks such as skc status, skc deliver --check, and skc update --check do none of these.