Agent Skill for Apple Swift Playgrounds
Giving Codex new capabilities and expertise - in iPad mini-app authoring
Introduction
OpenAI’s LLMs have improved utility in Swift coding a lot since GPT-4. But still, the Apple Swift Playground(s) variety - particularly on iPad - is not covered well. (Swift Playgrounds is positioned for programming beginners, but another plus is that it’s a dynamic code execution environment on iPad). With a recent EPUB creation project from online sources, the process degraded into the familiar back and forth with the AI implementing possible remedies to errors and me feeding iPad screenshots of new errors back to it.
The mistakes were basic ones, for example:
Manifest mismatches for Playgrounds app packages (.iOSApplication schema issues)
Missing AppleProductTypes import in Package.swift
Orientation/device-family enum references failing type inference
“Fixes” that worked in theory but not in the actual iPad runtime
In short: solid Swift generation, weak platform-specific operational reliability.
Agent Skills to the rescue:
Agent Skills are folders of instructions, scripts, and resources that agents can discover and use to do things more accurately and efficiently.
Swift Playgrounds App Builder Skill
The new Skill focuses on producing importable, runnable .swiftpm app bundles for iPad Swift Playgrounds without LLM guesswork.
Starting optionally from a known-good Playgrounds app structure (Package.swift, App.swift, optional Workspace.plist), it establishes a playbook from lessons learned:
Use app-style entrypoint (@main, WindowGroup) only
Prefer Playgrounds-safe sharing (ShareLink) over UIKit wrappers by default
Pre-produce artwork assets on Mac if support on iPad is uncertain
Validate package shape before handoff
Produce native .swiftpm bundle ready for AirDrop or iCloud sharing; not some other “build artifact”
Assume caching weirdness on iPad and force freshness when needed
Why a SKILL instead of just better prompting
What OpenAI describe as “Give Codex new capabilities and expertise”, the official Agent Skills website breaks down into four causes:
Domain expertise: Package specialized knowledge into reusable instructions.
New capabilities: Give agents new capabilities.
Repeatable workflows: Turn multi-step tasks into consistent and auditable workflows.
Interoperability: Reuse the same skill across different skills-compatible agent products.
Translated to the realm of Apple Swift Playgrounds, the Skill builds on the LLM’s Swift generation abilities and extends these through specific knowledge about the restrictions on iPad, tells it about tools and methods to use, provides troubleshooting guidance - and hopefully will also help non-Codex users.
With that, the prompt to create a working Hello-World app becomes straightforward.
