Load UI, font, and documentation skills on demand

Who it's for

If you are already building interfaces and documents and want an agent to fill in practices beyond ordinary code review, this skill pack is a good fit. The official description covers UI audits, fonts, documentation, pull request reviews, and releases, loaded on demand by coding agents. There are twenty-five skills in total, covering loading states, type scales, and whether half of a diff consists of template-generated content.

The targets in the installation command are claude-code, codex, and cursor. That is the list written by this command; it does not mean the pack can only be used with those three, nor does it provide separate installers for other products.

It is best suited to people using Claude Code, Codex, or Cursor locally who want to start with a read-only UI or font review rather than changing releases immediately. See the official description at

https://github.com/mblode/agent-skills

First try in 3 minutes

Use only this official installation path, and separate the names after --agent with spaces:

npx skills add mblode/agent-skills -g --agent claude-code codex cursor -y

The official instructions state that comma-separated values are treated as one invalid name and rejected. Do not rewrite this as a comma-separated list.

Prerequisite: use one of the agents above locally. Cloud agents run in a fresh virtual machine and cannot see ~/.claude/skills or ~/.agents/skills on your machine; Cursor's Sync Skills only covers ~/.cursor/skills. If you try this in a cloud session, a global installation on your machine will not appear automatically.

Choose a read-only check for the first task. Open a real page or component and ask the agent to audit the interface, focusing on loading states; or ask it to check the type scale on the page. The skills load on demand, and the page does not list a slash command for every skill. Do not first ask it to change releases or directly finalize documentation.

Success signal: the installation command accepts the space-separated --agent values, after which the agent loads the relevant skills for the task and gives specific feedback on loading states or type scale, instead of reporting an invalid agent name or saying that the skill cannot be found in the cloud.

Use cases

Case one: start with a UI audit focused only on loading states.

The official description lists UI audits among the capabilities and uses loading states as an example of the gap these twenty-five skills are meant to cover. For the first pass, request feedback only; do not change code. The page does not provide individual checklist headings or say that a particular framework must be opened. What it does establish is that loading states are in scope and that the skills load on demand.

Case two: check the type scale first, then decide whether to change the styles.

The official description also names fonts and uses type scales as an example. For the first pass, ask the agent only to compare the size relationship between the existing headings and body text and judge whether the scale is inconsistent. Do not write a stylesheet until you confirm. The page does not provide a recommended size table or a font file path.

Pull request reviews and releases appear in the same sentence, but do not take an action that changes version records or merge notes on the first pass. The page does not expand on the steps for either capability.

Boundaries

--agent accepts only space-separated values. Writing a comma-separated list fails as a whole. Cloud sessions cannot see ~/.claude/skills or ~/.agents/skills; Cursor sync only covers ~/.cursor/skills. The page does not explain how to inject local skills into a cloud virtual machine.

The capabilities are summarized as UI audits, fonts, documentation, pull request reviews, and releases, with an emphasis that these are areas ordinary code review often misses. The page does not give the individual command names for all twenty-five skills, nor does it specify language or framework restrictions.

You need to confirm before turning the feedback into a commit. The page does not promise automatic merging or automatic releases.