before skills

Agent skills from before.click, the App Store design gallery.

A skill is a set of instructions your coding agent loads when it recognises the task. Instead of improvising an answer about App Store listings, it follows a fixed process, with rules and reasoning written down in advance.

npx skills add alexszczurek/before-skills

Works with Claude Code, Cursor, and any agent that reads SKILL.md files. The installer asks where to put the skill: globally, so it is available in every project, or in the current project only.

To install one skill from this repo:

npx skills add alexszczurek/before-skills --skill aso

Writes and audits iOS App Store listings: app name, subtitle, keyword field, screenshots, icon, preview video, tags, ratings, localisation, and featuring nominations.

Distilled from the before.click ASO guide and checked against Apple's 2025–2026 documentation, so it covers what changed since most ASO advice was written: Apple-confirmed screenshot indexing, AI-generated App Store tags, Liquid Glass icons, the shift toward semantic search, and the App Store Review Guideline that prohibits competitor names in the keyword field.

Ask for what you want in plain language. The skill loads on its own when your request mentions App Store metadata, keywords, screenshots, ratings, or ASO:

• "Audit my App Store listing" — walks the audit checklist in order, cheapest fix first
• "Write a title and subtitle for my caffeine tracking app" — runs the keyword research procedure, then distributes terms across fields
• "Is my keyword field any good?" — checks character count, formatting, duplicate words, and prohibited terms
• "Plan my next metadata update" — separates what needs an app release from what you can change today

In Claude Code you can also invoke it directly with /aso.

Ask an agent about ASO without a skill and you get a different answer each time, some of it out of date. The skill fixes the process, not the answer.

It follows a procedure instead of guessing. Keyword research runs in six steps: brainstorm, check App Store autocomplete, extract competitor terms, score for volume and relevance and difficulty, group by search intent, then distribute across fields. It cannot skip to writing metadata from intuition.

Every rule carries its reason. "Singular forms only, because Apple matches plurals automatically, so the plural wastes characters." The reasoning lets the agent extend a rule to situations the skill never described.

Constraints are numbers, not adjectives. Fill at least 95 of 100 keyword characters. Three system rating prompts per year, total. Submit featuring nominations three weeks ahead. Judge a metadata change at two to four weeks, never the next day.

It knows what it cannot prove. Claims are marked as Apple-confirmed or industry inference, so the agent does not present a guess as documentation.

Open an issue with the source, and the rule and its reasoning. Rules need evidence: Apple documentation, or data from a reputable ASO source.

MIT