v3.3.0

Added

• Publish the scoped Simulator runtime with 14 MCP tools: installed Xcode/device discovery, native Simulator display, private sidebar preview, build/test/install/launch and screenshots with optional MCP image output.
• Add a loopback browser viewer with token-protected screenshot access, bounded refresh, pause/fit controls and cleanup on stop. Native Simulator handles touch/keyboard interaction.
• Add dated iPhone Duo guidance grounded in Apple's September 9 announcement. Discover actual installed profiles rather than inventing Duo support; this development Mac has no Duo device type.
• Refresh Apple technology and updates snapshots; bundle the new local guidance in MCP 2.4.0 and integration archives.

---

Full changelog: https://github.com/Nagarjuna2997/ios-agent-skill/blob/v3.3.0/CHANGELOG.md

v3.2.1

Added

• Reduce the always-loaded skill entry point and move the full engineering rules into an on-demand guide.
• Bundle canonical local guides, source files, templates and assets in the knowledge MCP; store identical bodies once by content hash.
• Add offline source search, heading outlines and bounded exact-content reads with continuation offsets. Technology lookup defaults to a compact overview; full guide retrieval remains explicit.
• Add an original AppleRecipes Swift package with source, tests and Apple guidance attribution. Keep implementation coverage separate from the 405-technology directory.
• Ship MCP 2.3.1 and updated ChatGPT/Codex/Gemini integration archives with the offline source workflow.

---

Full changelog: https://github.com/Nagarjuna2997/ios-agent-skill/blob/v3.2.1/CHANGELOG.md

v3.1.0

Added

• Complete 405-technology Apple directory, 96 update/release-note landing pages, and layer-by-layer Icon Composer guidance.
• ios-agent-mcp 2.2.0 includes a separate knowledge server with five tools, stdio and Streamable HTTP, bundled public reference data, and protocol tests.
• @nagarjuna2002/ios-agent 0.2.0 adds app briefs, optional XcodeGen scaffolding and editable SVG icon layers with overwrite protection.
• Portable ChatGPT skills-only and Codex plugin ZIPs, Gemini CLI extension, and explicit client setup instructions.
• GitHub Release assets include npm tarballs, self-contained plugin bundles and SHA-256 checksums. Public marketplace acceptance and a hosted endpoint are separate deployment steps.

---

Full changelog: https://github.com/Nagarjuna2997/ios-agent-skill/blob/v3.1.0/CHANGELOG.md

v2.0.1

Fixed

• ios-agent-mcp reported version 1.0.0 from --version and, more consequentially, from the MCP initialize handshake -- so every connected client saw a version that did not match the package it had just installed. The version was a hardcoded constant in mcp-server/src/index.ts and nothing read package.json at runtime, so publishing under a new version could not correct it on its own. The constant, mcp-server/package.json, skill.json, and the SKILL.md frontmatter are now aligned on a single version.

---

Full changelog: https://github.com/Nagarjuna2997/ios-agent-skill/blob/v2.0.1/CHANGELOG.md

v2.0.0

Added

Turns the repository from something you read into something you install. The skill teaches an agent how to write iOS code; the MCP server lets it check code that already exists.

• mcp-server/ -- ios-agent-mcp, a TypeScript MCP server (stdio transport) exposing six iOS-specific tools:
  • analyze_swift_project -- structure (file/line counts, deployment target, Swift tools version, frameworks, test presence) plus a finding count per category with the tool that explains each.
  • review_swift_concurrency -- @Observable without @MainActor, Task.detached, DispatchQueue.main.async, await MainActor.run inside an isolated type, @unchecked Sendable, nonisolated(unsafe), unstructured Task in onAppear, empty catch, non-final observable classes, and types named Task that shadow _Concurrency.Task.
  • review_swift_architecture -- initializers defaulting to live implementations, presentation code naming URLSession/APIClient/ModelContext, singletons resolved inside view models, the domain layer importing SwiftUI, nested NavigationStacks, and NavigationView.
  • review_swiftui -- fixed font sizes and heights, AnyView, deprecated .cornerRadius, literal spacing instead of tokens, materials over solid backgrounds, transient view state on models, ObservableObject, @EnvironmentObject, and try!.
  • check_availability_guards -- missing guards and over-restrictive ones: an iOS 26 API guarded at #available(iOS 27, *) compiles, ships, and silently drops every iOS 26 device to the fallback. Also flags Foundation Models used without a runtime SystemLanguageModel.availability check.
  • audit_app_store_readiness -- permission-gated frameworks with no Info.plist purpose string, a missing PrivacyInfo.xcprivacy (apps only -- libraries are never submitted), unlocalized user-facing strings, unlabeled icon-only buttons, and print() used for diagnostics.
• Every finding carries a file, a line, the severity, the consequence, the specific fix, and a link into this repo's docs. Analyzers are pure (path, content) -> Finding[] functions, so they are unit-testable without the MCP transport.
• 38 tests -- unit coverage for every analyzer plus test/server.smoke.test.js, which launches the real server and speaks the real MCP protocol over stdio. Unit tests cannot tell you whether the server actually starts; that one can.
• docs/mcp/ -- installation.md (Claude Code, Claude Desktop, Cursor, from source, troubleshooting, privacy), tools.md (every rule with its severity, and the limits of static analysis), examples.md (worked sessions, including how the tools pair with the subagents).
• mcp-server/mcp.json -- manifest declaring runtime, transport, filesystem-read-only permissions, and the tool list.
• CI gained an mcp-server job: install, typecheck, build, test, validate the manifest, and verify the manifest's tool list matches the tools actually registered in src/index.ts -- a tool advertised but not registered would be a broken promise to any client.

Fixed

• Dogfooding the server against this repo's own samples/SkillPatterns surfaced two false positives in the analyzers, both fixed with regression tests: empty-catch matched catch { } inside a doc comment because that one check used a raw regex instead of the comment-stripping line walker every other rule uses; and missing-privacy-manifest fired on an SPM library, which has no Info.plist and is never submitted to App Review.
• Package.swift is no longer analyzed as application source -- it is build configuration, and including it inflated file counts and produced findings against code that is not part of the app.
• A line field in the availability analyzer held the matched source string rather than the line number; it type-checked only because of an unsafe as unknown as cast. Fixed, cast removed, regression test added.

---

Full changelog: https://github.com/Nagarjuna2997/ios-agent-skill/blob/v2.0.0/CHANGELOG.md

v1.0.0 — Stable iOS Agent Skill

Stable iOS Agent Skill

This release marks the first stable version of the iOS Agent Skill.

It includes production-oriented guidance for AI coding assistants working with iOS, Swift, SwiftUI, Xcode project workflows, Apple platform conventions, testing expectations, and multi-agent instruction files.

Included

• Core Agent Skill instructions in SKILL.md
• Guidance for Claude Code, Codex, Cursor, GitHub Copilot, Gemini CLI, Windsurf, and other coding agents
• Xcode project-first workflow documentation
• Swift, SwiftUI, architecture, testing, and platform guidance
• Adoption evidence tracking through ADOPTION.md

This release is intended as a stable baseline for community usage and future curated-list submissions.