Overview

How might we…give someone a focused, repeatable way to audit a Swift app with Claude Code, without writing each prompt from scratch?

Problem

A Swift agent skill knows what good Swift looks like, but /swiftui-pro on its own does not say what to audit, where to look or what to hand back. Someone has to write that prompt for every area of an app, from concurrency and security to accessibility and interface copy, and it has to ask for something you can act on.

Twenty-two audits in seven waves

Each audit is a prompt for a fresh Claude Code session. Its first line is the slash command for the skill that does the work, and the rest names what to read, what to check and what to hand back: findings by file and line, each with a fix, so you get a list you can triage. A few ask for a plan, a checklist or a recommendation instead.

The waves set the order. A-01 sets the baseline, A-02 to A-04 cover foundational correctness, and the later waves go through performance and accessibility, modernization and UI quality, code quality and testing, and specialist audits. Two multi-agent reviews run last because they send several agents across the whole codebase. Each audit has a size from S to XL and a rough time that includes triaging what it finds.

Fill in your project once

Four fields at the top fill in every prompt: app name, project path, iOS deployment target and Swift version. Backend and Figma file are optional, and when one is blank the lines that mention it drop out of the prompt. A blank iOS or Swift field defaults to 26 or 6.2. The highlighted text in each prompt shows what was filled in.

Fill in your project once and every prompt updates. Audit A-03 is open, with the app name, project path and backend highlighted where they were filled in.

Copy, paste, triage

Copy an audit into a fresh Claude Code session in your project root and let it run. Mark it done once you have triaged what it found: decide what to fix now, later or not at all. A progress bar and the To do and Done filters keep track, and the progress is saved only in your browser. An optional base prompt loads all 20 Swift skills in one message first, and Playlist, another project in this portfolio, can save that list as one slash command.

Decisions that shaped it

Credit the skills, do not copy them

The prompts use 27 skills written by 17 other authors. The page links to each one's own repository instead of copying it, and each keeps its own license: 25 are MIT, one is Apache 2.0, and one has no license file, so the page says to install that one from its own repo.

One file, no network

The page is a single index.html with no dependencies, no build step and no network requests. Its Content Security Policy blocks anything that tries, there is no analytics, and your project details and progress never leave your browser. You can download it and open it offline, or fork it and edit the prompts for your team.

A check that keeps the page honest

A short script checks that the prompts, the skills list, the Content Security Policy hashes and the counts quoted in the README all agree, and a workflow runs it on every change to main and every pull request. Adding an audit means copying one entry and running the script, which lists any count that was missed.

Plain words and a next step

The page's own copy follows a short style guide: sentence case titles that say what an audit checks, the same word for the same thing everywhere (audit, prompt, base prompt, skill, triage), and a message after every action that names the next step.

Shipping it

Open source under the MIT license, published in October 2026 with the page live on GitHub Pages, issue templates, a code of conduct and a security policy. New audits and sharper wording are welcome. It is not affiliated with Apple or Anthropic.

github.com/mvoshell652/swift-audit-prompts