Post

@hishnash.bsky.social & @natpanferova.bsky.social turned their books into agent skills, but agents didn't pay enough attention to them. So they built SwiftFairy: a local macOS static code analyser with an MCP server! go.peterfriese.dev/ai-swiftfai...

tools.nilcoalescing.comSwiftFairyA macOS app that locally reviews your agent's Swift and SwiftUI code and provides guidance to fix issues.

15/10/26, 4:30 pm2 reposts10 likes

Axialis

@axialis.com

Turning guidance into a local check is key; agents can ignore prose, not a failing analyzer.

anhphong

@anhphong.bsky.social

Matches what I kept measuring with my own agent rules: prose guidance competes for attention with everything else in the context, so a new rule can dilute the ones already not holding. I had one violated three minutes after writing it. A static gate that fails the build doesn't compete.

Scott Shapiro

@pixelscott.bsky.social

turning the book into a local analyser with an mcp server is the right escalation. agent skills are documentation, an mcp server is capability. big difference.

Peter Friese

@peterfriese.dev

I've been experimenting with MCP servers for coding agents a lot lately. You can read more about how SwiftFairy fits into that in Not only Swift #106: go.peterfriese.dev/ai-swiftfai...

peterfriese.devNot only Swift - Issue #106How native SwiftUI controls adopt Liquid Glass in iOS 26 without extra code. Plus: Cupertino.tech's Apple SDK timeline, Margaret Hamilton's pioneering legacy, SwiftFairy code review for AI agents, iterative data loading in Swift, and reviving dinner parties.