agentleFS
kelvinkosbabkelvinkosbab/KozBon/CLAUDE.md

KozBon

CLAUDE.md5 starsChanged 10 days agoMITView on GitHub

KozBon is a multi-platform Apple app for discovering and managing Bonjour (mDNS) network services. Bundle ID: com.kozinga.KozBon. Execution conventions (plan multi-step edits, no self-narration, atomic edits) live in .claude/rules/claude-execution-discipline.md, which loads every session. Never push to main. Make every change on a branch and open a pull request against main. CI runs on the PR. The KozBon scheme has no test action configured; all tests run through SPM. KozBon.xcworkspace contains: - KozBon.xcodeproj — the app target (2 Swift files) -…

The file

We split this CLAUDE.md into the rules it gives an agent

7 rules. About 0 are common, 0 are distinctive, and 7 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈0

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 7

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Use // MARK: - section headers to organize code within files

    line 212
  2. Very specific:

    Use the zero-parameter onChange(of:) closure (iOS 17+), not the deprecated { _ in } form

    line 214
  3. Very specific:

    Service descriptions: use serviceType.localizedDetail, not .detail.

    line 234
  4. Very specific:

    Adding strings: add the key to Strings.swift and the entry with all 8 translations to Localizable.xcstrings.

    line 236
  5. Very specific:

    Verify a change by building and inspecting Metadata.appintents/extract.actionsdata in the built app — autoShortcuts should list both …

    line 256
  6. Very specific:

    All workflows use macos-26 runner with concurrency groups and SPM caching.

    line 280
  7. Very specific:

    Info.plist NSBonjourServices: Every service type the app wants to discover must be listed here.

    line 285

≈ “Common” is an estimate: at least 1 in 5 of the 27,641 CLAUDE.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
CLAUDE.md

CLAUDE.md

Project Overview

KozBon is a multi-platform Apple app for discovering and managing Bonjour (mDNS) network services. Bundle ID: com.kozinga.KozBon.

Working with Claude

Execution conventions (plan multi-step edits, no self-narration, atomic edits) live in .claude/rules/claude-execution-discipline.md, which loads every session.

Never push to main. Make every change on a branch and open a pull request against main. CI runs on the PR.

Build & Run

# Build for iOS Simulator
xcodebuild -workspace KozBon.xcworkspace -scheme KozBon -destination 'platform=iOS Simulator,name=iPhone 17 Pro' build

# Build for macOS
xcodebuild -workspace KozBon.xcworkspace -scheme KozBon -destination 'platform=macOS' build

# Build for visionOS Simulator
xcodebuild -workspace KozBon.xcworkspace -scheme KozBon -destination 'platform=visionOS Simulator,name=Apple Vision Pro' build

# Run tests — all tests live in the SPM package
swift test --package-path KozBonPackages

The KozBon scheme has no test action configured; all tests run through SPM.

Architecture

  • Swift 6.2 with strict concurrency checking (SWIFT_STRICT_CONCURRENCY = complete)
  • SwiftUI with MVVM pattern, targeting 26 on every platform (iOS, macOS, tvOS, watchOS, visionOS). The minimums were raised together so that no #available gating is needed for 26-era APIs — Liquid Glass, Foundation Models, and the Apple Intelligence surfaces are all unconditional. Only two availability annotations remain in the codebase: an iOS 27 / visionOS 27 check for TabRole.prominent in AppCoreScene.swift, and one @available(macOS, unavailable), which is a platform exclusion rather than a version gate. Adding a new #available(… 26 …) is a sign something is out of date.
    • View-model conventions are documented in .claude/rules/apple-swiftui-mvvm.md — when to use a VM, @State vs @Bindable ownership, dependency-plumbing rules, splitting large VMs across companion files
  • Modular SPM packages via KozBonPackages/ local package in the Xcode workspace
  • Dependency Injection via DependencyContainer (in BonjourScanning module) using SwiftUI environment (@Environment(\.dependencies))
  • Core Data for persistence (iDiscover.xcdatamodeld) — all Core Data access is @MainActor via viewContext
  • Localization via BonjourLocalization module with String Catalog (.xcstrings) — 8 languages supported, including right-to-left (Arabic, Hebrew)
  • Protocol-based abstractions: BonjourServiceScannerProtocol, BonjourPublishManagerProtocol

Project Structure

Workspace

KozBon.xcworkspace contains:

  • KozBon.xcodeproj — the app target (2 Swift files)
  • KozBonPackages/ — local SPM package (16 modules)

App Target (KozBon/)

Deliberately thin — the entry point, the one type the App Intents extractor requires here, and the resources that must live in the main bundle:

  • KozBonApp.swift — @main entry point; pulls the root scene from AppCore
  • KozBonAppShortcuts.swift — AppShortcutsProvider. Cannot move into a package; see "App Intents & Siri Strings"
  • Localizable.xcstrings — App Intents strings (main-bundle-only)
  • AppShortcuts.xcstrings — spoken Siri phrases
  • InfoPlist.xcstrings — permission usage descriptions
  • Info.plist, PrivacyInfo.xcprivacy, Assets.xcassets, service-names-port-numbers.csv

SPM Modules (KozBonPackages/)

Each module follows the {name}/Sources and {name}/Tests layout:

ModulePurposeKey Types
AppCoreRoot scene, its view model, macOS menu commands, top-level tab destinationsAppCoreScene, AppCoreViewModel, TopLevelDestination
LocalNetworkMonitorNWPathMonitor-backed Wi-Fi / Ethernet path check so Discover can explain why a scan can't reach anythingLocalNetworkMonitor, its @MainActor protocol + mock
InternetConnectivityNWPathMonitor-backed internet reachability, the ConnectivityRequirement features declare, and URLError offline classification. Gates the cloud AI backends; Bonjour and Apple Intelligence never wait on itInternetConnectivityMonitor, its @MainActor @Observable protocol + mock, ConnectivityRequirement, ConnectivityFailure, EnvironmentValues.internetConnectivity
BonjourCoreValue types, constants, utilitiesConstants, TransportLayer, InternetAddress, Logger, Clipboard
BonjourStoragePersistence — SwiftData preferences + Core Data custom-service-type storePreferencesStore, UserPreferences, CustomServiceType, MyCoreDataStack, MyDataManagerObject
BonjourLocalizationLocalized strings (8 languages, including RTL: ar, he)Strings enum, Localizable.xcstrings
BonjourModelsDomain models and service libraryBonjourServiceType, BonjourService, BonjourServiceSortType
BonjourScanningNetwork discovery and publishingBonjourServiceScanner, MyBonjourPublishManager, DependencyContainer, mocks
BonjourAICoreProvider-agnostic AI scaffolding — protocols, value types, prompt builders, safety, mocks, simulator stubs, UI primitives, credentials-store protocol + Keychain/InMemory implsBonjourChatSessionProtocol, BonjourServiceExplainerProtocol, BonjourChatPromptBuilder, BonjourServicePromptBuilder, AIBackend, AICloudProvider, AICloudCredentialsStore, AICloudError, KeychainAICloudCredentialsStore, InMemoryAICloudCredentialsStore, MockBonjourChatSession, MockBonjourServiceExplainer, ServiceExplanationSheet, MarkdownContentView, TypingIndicator
BonjourAIAppleApple Foundation Models implementations of the BonjourAICore protocolsBonjourChatSession, BonjourChatSessionFactory, BonjourServiceExplainer, BonjourServiceExplainerFactory, AppleIntelligenceSupport, AIContextMenuItems, prepare-tool wrappers
BonjourAIAnthropicAnthropic Claude implementations of the BonjourAICore protocols + the runtime model catalog and the PreferencesStore.aiCloudModelIdentifier bridgeAnthropicModel (offline fallback only), AnthropicModelOption, AnthropicModelCatalog, AnthropicModelCatalogClient, AnthropicClient, AnthropicConfiguration, AnthropicBonjourChatSession, AnthropicBonjourServiceExplainer, MockAnthropicClient
BonjourAIGeminiGoogle Gemini implementations of the BonjourAICore protocols + the runtime model catalog. Uses the Gemini Developer API (generativelanguage.googleapis.com), not Vertex AI — Vertex needs service-account OAuth the paste-an-API-key sheet can't express.GeminiModel (offline fallback only), GeminiModelOption, GeminiModelCatalog, GeminiModelCatalogClient, GeminiClient, GeminiConfiguration, GeminiGenerateRequest, GeminiStreamEvent, GeminiBonjourChatSession, GeminiBonjourServiceExplainer, MockGeminiClient
BonjourAIOpenAIOpenAI implementations of the BonjourAICore protocols + the runtime model catalog. Uses the Responses API (/v1/responses) with store: false, so OpenAI keeps no copy of the conversation.OpenAIModel (offline fallback only), OpenAIModelOption, OpenAIModelCatalog, OpenAIModelCatalogClient, OpenAIClient, OpenAIConfiguration, OpenAIResponseRequest, OpenAIStreamEvent, OpenAIBonjourChatSession, OpenAIBonjourServiceExplainer, MockOpenAIClient
BonjourAIUmbrella module — cloud-aware routing factories + @_exported import BonjourAICore so legacy import BonjourAI consumers stay workingCloudAwareBonjourChatSessionFactory, CloudAwareBonjourServiceExplainerFactory
BonjourAppIntentsThe Siri / Shortcuts / Spotlight actions. Holds the intents and entity; the AppShortcutsProvider itself lives in the app target because Xcode won't extract or localize one declared in a packageScanForServicesIntent, ListDiscoveredServicesIntent, FindServicesIntent, IsDeviceOnlineIntent, OpenServiceIntent, ExplainServiceTypeIntent, BroadcastServiceIntent, StopBroadcastingIntent, BonjourServiceEntity, BonjourServiceTypeEntity, ServiceCategoryAppEnum, AppIntentsBridge, RecentServiceEntities, SiriNetworkSummary, IntentScan
BonjourUISwiftUI views and view modelsAll views, BonjourServicesViewModel, UI components

Dependency Graph

App → AppCore, BonjourAppIntents
AppCore → BonjourUI, BonjourAppIntents, BonjourAI, BonjourAIApple, BonjourAIAnthropic, BonjourAIGemini, BonjourAIOpenAI, BonjourScanning, BonjourModels, BonjourStorage, BonjourLocalization, BonjourCore, InternetConnectivity, Core, CoreUI
BonjourAppIntents → BonjourAI, BonjourModels, BonjourScanning, BonjourLocalization, LocalNetworkMonitor, InternetConnectivity, BonjourCore
BonjourUI → BonjourModels, BonjourScanning, BonjourLocalization, BonjourAI, BonjourAIApple, BonjourAIAnthropic, BonjourAIGemini, BonjourAIOpenAI, BonjourStorage, BonjourCore, InternetConnectivity, Core, CoreUI
BonjourAI → BonjourAICore, BonjourAIApple, BonjourAIAnthropic, BonjourAIGemini, BonjourAIOpenAI, BonjourCore, BonjourModels, BonjourLocalization, BonjourScanning, BonjourStorage
BonjourAIApple → BonjourAICore, BonjourCore, BonjourModels, BonjourLocalization, BonjourScanning, BonjourStorage
BonjourAIAnthropic → BonjourAICore, InternetConnectivity, BonjourCore, BonjourModels, BonjourLocalization, BonjourScanning, BonjourStorage
BonjourAIGemini → BonjourAICore, InternetConnectivity, BonjourCore, BonjourModels, BonjourLocalization, BonjourScanning, BonjourStorage
BonjourAIOpenAI → BonjourAICore, InternetConnectivity, BonjourCore, BonjourModels, BonjourLocalization, BonjourScanning, BonjourStorage
BonjourAICore → InternetConnectivity, BonjourCore, BonjourModels, BonjourLocalization, BonjourScanning, BonjourStorage
BonjourScanning → BonjourCore, BonjourModels, LocalNetworkMonitor
BonjourModels → BonjourCore, BonjourStorage, BonjourLocalization
BonjourStorage → BonjourCore
BonjourLocalization → (Foundation only)
LocalNetworkMonitor → Core
InternetConnectivity → Core
BonjourCore → Core (BasicSwiftUtilities)

AI Backend Routing

ADR 0005 introduces a pluggable AI backend (extended by ADRs 0006 and 0007). The Settings → AI Backend section exposes a picker between four options:

  • Apple Intelligence (default) — on-device via BonjourAIApple and FoundationModels.
  • Anthropic Claude (opt-in) — cloud via BonjourAIAnthropic and the user's own API key.
  • Google Gemini (opt-in) — cloud via BonjourAIGemini and the user's own Google AI Studio key.
  • OpenAI GPT (opt-in) — cloud via BonjourAIOpenAI and the user's own OpenAI API key. Labelled "OpenAI GPT", not "ChatGPT": a ChatGPT subscription doesn't include API access, and the sign-in copy says so. GitHub Models was a third option until GitHub retired the service on 2026-07-30. AIBackend.github and the whole BonjourAIGitHub module are gone; a stored "github" preference migrates to Apple Intelligence via AIBackend.resolved(rawValue:), and Settings shows a one-time notice offering to delete the orphaned Personal Access Token (the presence of that Keychain entry is the notice's only state). AICloudProvider.github and Color.kozBonGitHub deliberately survive — they are what that cleanup row is built from — so don't "finish the job" by deleting them.

CloudAwareBonjourChatSessionFactory / CloudAwareBonjourServiceExplainerFactory live in the BonjourAI umbrella and sit above the per-provider factories in BonjourAIApple / BonjourAIAnthropic / BonjourAIGemini / BonjourAIOpenAI. They read preferencesStore.aiBackend on every makeForCurrentEnvironment(...) call and route to the right implementation. AppCoreScene watches preferencesStore.aiBackend and aiCloudModelIdentifier via .onChange and calls AppCoreViewModel.refreshAIBackend() so flipping the picker takes effect without an app restart (the in-flight conversation is dropped across the swap).

Each provider's API key lives in the iOS Keychain (whenUnlockedThisDeviceOnly, never iCloud-synced) via KeychainAICloudCredentialsStore, keyed by the provider's raw value. Tests substitute InMemoryAICloudCredentialsStore.

Gemini keys go in the x-goog-api-key header, never the ?key= query parameter the Gemini docs also accept — query strings land in logs, proxies, and crash reports, and this is a user credential.

The Claude model picker is populated at runtime by AnthropicModelCatalog (GET /v1/models with the user's own key), not by the AnthropicModel enum — that enum is now only the offline fallback. Resolution is three-tiered: live fetch → this session's cached fetch (1h TTL, in-memory) → the compiled-in list. BonjourAIGemini mirrors this with GeminiModelCatalog against GET /v1beta/models, filtered to models advertising generateContent — the endpoint also returns embedding, image, and TTS variants that a chat picker must not offer. OpenAIModelCatalog does the same against GET /v1/models, which advertises no capabilities at all, so its filter is by identifier family (OpenAIModelCatalogClient.isChatModel) — the piece to revisit when OpenAI adds a model family.

AIBackendAvailability (in BonjourAICore) offers only Apple Intelligence on the China mainland App Store storefront (Storefront.countryCode == "CHN"): App Review rejected 4.8 under Guideline 5 because ChatGPT-associated features need a permit there, and the same rule covers every unfiled generative-AI provider. Release notes that name a cloud provider (namesCloudProvider) are hidden from What's New and the chat prompt there too. Settings lists only offeredBackends, both routing factories take an isBackendOffered closure that also blocks fallbacks, and a storefront change resets a withheld selection to the default. An unknown storefront offers everything. App Store metadata in every locale names no cloud AI provider — see docs/app-store-metadata.md.

OpenAI's reasoning models spend the output budget on hidden reasoning, so the OpenAI session caps turns at 4,096 tokens (not 1,024) and sends reasoning.effort: "low" only to families known to accept it (OpenAIReasoning.forModel) — an unsupported parameter fails the request.

The selected model persists per provider: aiCloudModelRawValue holds the Claude choice, aiGeminiModelRawValue the Gemini one, and aiOpenAIModelRawValue the OpenAI one, so switching backends and back preserves each rather than handing Gemini a claude- identifier. Read them through PreferencesStore.aiCloudModelIdentifier(for:) in BonjourAICore (the no-argument property resolves the currently-selected backend). Values are raw Strings, and an identifier the binary doesn't recognize is only discarded when a live list proves it retired — otherwise a model newer than the build would be silently replaced by the default.

Offline Handling

Features declare a ConnectivityRequirement: .internet for every cloud chat session and explainer (and AIBackend.connectivityRequirement for cloud backends), .none for Apple Intelligence, mocks, and simulator stubs. Gate on the session's requirement, not the selected backend, because the routing factories can fall back across backends. AppCoreViewModel owns the live InternetConnectivityMonitor, starts it at init (Siri may run with no scene), injects it as \.internetConnectivity, and hands AppIntentsBridge an isInternetConnected closure.

The monitor is @Observable, so the UI follows connectivity live rather than asking the user to retry: the chat shows ChatOfflineBanner and disables Send, the Insights sheet runs its .task(id:) once the connection returns, and Settings defers the model-catalog fetch the same way. Requests that still fail for lack of a connection show AICloudError.displayMessage(for:), which localizes .networkUnavailable instead of surfacing its English log text. LocalNetworkMonitor answers a different question (is the local link up for Bonjour?) and stays separate.

Siri Actions and App State

Most Siri actions scan with their own IntentScan and need nothing from the app. The ones that act on app state (Broadcast, Stop Broadcasting, Explain, Open in KozBon) go through AppIntentsBridge.shared, which AppCoreViewModel.connect(to:) fills in. That happens in AppLaunch, reached from KozBonApp.init() via AppCoreScene.prepareForLaunch(), because the system can launch the app in the background for an action and never build a scene. Keep that wiring out of the scene.

Code Conventions

  • Use // MARK: - section headers to organize code within files
  • One type per file, feature-based folder organization
  • Use the zero-parameter onChange(of:) closure (iOS 17+), not the deprecated { _ in } form
  • All user-facing strings must use BonjourLocalization.Strings.* — never hardcode English strings in views. Two compiler-enforced exceptions, both in the app target (see "App Intents & Siri Strings" below): App Intents metadata and App Shortcut phrases cannot resolve against a package bundle. What Siri says at runtime is not metadata, so it goes through Strings.Siri like any other string.
  • Service type definitions go in MyServiceType+Library.swift — each new type needs a static private let definition AND an entry in tcpServiceTypes or udpServiceTypes array AND a corresponding NSBonjourServices entry in Info.plist

Concurrency (project specifics)

Swift 6 language mode with complete checking; the general rules are in .claude/rules/apple-swift6-strict-concurrency.md. KozBon's specifics:

  • Service classes (scanners, publishers) are @MainActor final class. Core Data types (MyCoreDataStack, MyDataManager, CustomServiceType) are @MainActor, since all access goes through viewContext.
  • DependencyContainer is final class: Sendable with a @MainActor init() for production and a nonisolated init for tests.
  • @MainActor protocols (BonjourServiceScannerProtocol, BonjourPublishManagerProtocol) inherit AnyObject, Sendable.
  • NetServiceDelegate / NetServiceBrowserDelegate conformances use @preconcurrency, not nonisolated methods. When both sides of a delegate chain are @MainActor, call delegate methods directly.
  • nonisolated(unsafe) only for statics like static let sortDescriptors: [NSSortDescriptor]? = nil.

Localization

General rules: .claude/rules/apple-localization-best-practices.md. KozBon's specifics:

  • Module: BonjourLocalization, String Catalog Localizable.xcstrings, 8 languages: en (base), es, fr, de, ja, zh-Hans, ar, he. Arabic and Hebrew are right-to-left; mirror direction-dependent symbols such as the chat cards' arrow.up.right with .flipsForRightToLeftLayoutDirection(true).
  • Access: Text(Strings.NavigationTitles.nearbyServices), String(localized: Strings.Errors.portMin), and format functions like Strings.Errors.portMin(value).
  • Service descriptions: use serviceType.localizedDetail, not .detail.
  • Release notes: each release in ReleaseNotes.all has a 1–2 paragraph English summary, which is its own catalog key, shown on What's New via Strings.Settings.releaseSummary(_:). A new or reworded summary needs all 8 locales. The highlights bullets are English-only chat context and aren't translated.
  • Adding strings: add the key to Strings.swift and the entry with all 8 translations to Localizable.xcstrings. scripts/validate-localizations.py enforces completeness in CI.

App Intents & Siri Strings

App Intents strings are the one place the BonjourLocalization.Strings.* rule does not apply. The compiler enforces this — the constraints below are build errors, not style preferences.

They must resolve against the main bundle. Routing an intent's title through Strings fails with "AppIntents requires LocalizedStringResource to use the main bundle", because Strings accessors carry bundle: .atURL(Bundle.module.bundleURL). So these strings live in the app target's own catalogs, not in BonjourLocalization:

CatalogHoldsReferenced from
KozBon/Localizable.xcstringsintent titles, descriptions, Spotlight keywords, entity type name, shortcut tile shortTitles.init("intent_…") in BonjourAppIntents
KozBon/AppShortcuts.xcstringsthe spoken Siri phrases onlyextracted from KozBonAppShortcuts.swift
KozBon/InfoPlist.xcstringspermission usage descriptionsread by iOS directly

KozBonAppShortcuts must stay in the app target. Xcode passes --no-app-shortcuts-localization to the App Intents extractor for every SPM target and omits it only for the app target. Declared in a package, the phrases can't be localized and never reach Metadata.appintents at all — autoShortcuts comes out empty and Siri offers nothing.

Some parameters accept only a literal. title takes .init("key"), but TypeDisplayRepresentation(numericFormat:) and AppShortcut(shortTitle:) reject it — "must be initialized directly with a String literal." Those keep the literal in Swift, and the catalog entry is keyed on the literal's own text (e.g. "%lld Bonjour services", "Scan for Services").

Don't put shortTitle in AppShortcuts.xcstrings. That catalog accepts spoken phrases only; anything else draws "This phrase is not used in any App Shortcut" once per locale.

Verify a change by building and inspecting Metadata.appintents/extract.actionsdata in the built app — autoShortcuts should list both shortcuts with their phraseTemplates.

SwiftLint

  • Config: .swiftlint.yml (line length warns at 140; todo disabled). There is no Xcode build phase: CI (swiftlint.yml) runs swiftlint lint --strict on PRs that touch .swift files.
  • Run locally: swiftlint lint from the repo root.
  • Intentional exceptions use // swiftlint:disable:next <rule> with a reason (e.g. force_unwrapping in Core Data, force_cast in MyDataManagerObject). MyServiceType+Library.swift disables rules at file level.

Testing

  • Framework: Swift Testing (@Test, @Suite, #expect). General rules: .claude/rules/apple-testing-strategy.md.
  • Runner: swift test --package-path KozBonPackages is the only runner; every test lives in KozBonPackages/.
  • CustomServiceTypeTests in BonjourStorage need Xcode to compile .xcdatamodeld, so they're excluded from the SPM test target (testExcludes). The SwiftData PreferencesStore tests run under swift test.
  • Flaky tests (Xcode 27+): swift test --package-path KozBonPackages --maximum-repetitions 5 --repeat-until fail, scoped with --filter.

CI / GitHub Actions

WorkflowTriggerWhat it does
Native CI (ios.yml)Push + PR to mainBuild-only on iOS Simulator (iPhone 17 Pro) and macOS — no xcodebuild test; tests run under spm-tests.yml
SPM Package Tests (spm-tests.yml)Push + PR when KozBonPackages/ changesswift build + swift test
SwiftLint (swiftlint.yml)PR when .swift files changeLint with inline PR annotations
Multi-platform Build (multiplatform-build.yml)Push + PR to mainmacOS + visionOS builds in parallel

All workflows use macos-26 runner with concurrency groups and SPM caching.

Important Gotchas

  • No duplicate service types: Each (type, transportLayer) pair must be unique in the library arrays. Duplicates cause redundant scanners and duplicate discovered services.
  • Info.plist NSBonjourServices: Every service type the app wants to discover must be listed here. Format: _type._transport (e.g., _http._tcp). No duplicates.
  • service-names-port-numbers.csv: IANA registry reference file bundled as a resource. Not parsed at runtime — service types are hardcoded in the Swift library.
  • BonjourService.id is logicalIdentity (name|type|domain), so it's stable across launches and dedups per-interface duplicates.
  • Core Data model in package: iDiscover.xcdatamodeld lives in BonjourStorage/Sources/Resources/ (alongside the SwiftData preferences container) and is loaded via Bundle.module. SPM CLI (swift test) cannot compile .xcdatamodeld — use Xcode for the Core Data tests.
  • @_exported import Core: BonjourCore/Sources/Exports.swift re-exports the Core package so downstream modules get Logger/Loggable without explicitly importing Core.
  • SF Symbol validation: All imageSystemName values in BonjourServiceType+UI.swift are verified valid SF Symbols. When adding new ones, validate with NSImage(systemSymbolName:accessibilityDescription:).

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.