dpearson2699/swift-ios-skills

Summary

A collection of 56 markdown-based 'agent skills' — structured documentation files (SKILL.md) targeting iOS 26+, Swift 6.2, and modern Apple frameworks like Liquid Glass, SwiftData, and Foundation Models. These files are designed to be loaded into AI coding assistants (Claude Code, Copilot, Codex, Cursor) to give them up-to-date iOS development context. There is no runnable Swift code in this repo — it is entirely a knowledge/prompt-engineering artifact.

Great for

iOS developers who want to give AI coding assistants accurate, non-deprecated iOS 26 / Swift 6.2 context — especially for newer APIs that training data won't cover well (Liquid Glass, AlarmKit, EnergyKit, Foundation Models, PermissionKit)

Easy wins

• +Add missing `references/` folders — about 6 skills (speech-recognition, codable-patterns, device-integrity, metrickit-diagnostics, app-clips, speech-recognition) have no references/ directory unlike their peers, suggesting those skills are less complete
• +Audit SKILL.md files for accuracy against actual iOS 26 beta SDK docs, since the entire value prop depends on correctness and there are zero automated checks
• +Add a CONTRIBUTING.md explaining the SKILL.md schema/format so contributors know what structure is expected (there is currently none)
• +Add a CI workflow that at minimum lints markdown and validates that each skill folder contains both SKILL.md and a references/ directory

Red flags

• !PolyForm Perimeter 1.0.0 license prohibits competing commercial use — contributors should know their improvements cannot be used to build a competing skills product; this is NOT a standard open-source license
• !Single commit in the entire repo history — the 'last_commit_at' of 2026-03-17 with commit_count=1 means this was bulk-uploaded in one shot, with no visible iteration or community input
• !iOS 26, Swift 6.2, AlarmKit, EnergyKit, PermissionKit — these are either unreleased or very recently released APIs at time of writing; there is no way to verify the SKILL.md content is accurate without access to beta SDKs, yet the repo makes strong accuracy claims
• !The 'Agent Skills' standard badge links to agentskills.io / skills.sh — these appear to be newly coined ecosystems with unclear adoption; the 'npx skills add' and '/plugin marketplace' commands may not work as advertised
• !No tests, no CI, no validation of any kind — for a repo whose entire value is the accuracy of its content, there is zero infrastructure to catch errors
• !218 stars with 1 commit and 0 issues suggests organic discovery is outpacing actual vetting of the content

Code quality

There is no source code to review — the entire repo is markdown documentation. The structure is clean and consistent: every skill lives in skills/<name>/SKILL.md with an optional references/ subfolder. The README is well-organized with bundle tables. However, with only 1 commit and 1 contributor, it's impossible to judge content depth or accuracy from the file tree alone, and the PolyForm Perimeter license (NOASSERTION in the API) actively restricts commercial use in ways many open-source contributors may not expect.

What makes it unique

The concept of framework-specific agent skill files is genuinely useful and underserved for iOS — most similar repos cover web/Python ecosystems. However, this repo's value is entirely dependent on content accuracy, and targeting unreleased/beta iOS 26 APIs makes it impossible to independently verify. It occupies a real niche but is essentially a content repo dressed up with ecosystem branding (agentskills.io) that has no proven traction.