CHANGELOG.md

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[Unreleased]

Removed — the System Extension subsystem

Some 24,000 lines that no shipping path could reach, and a release artifact no installed component could locate.

Activation was never attempted: both OSSystemExtensionManager.submitRequest call sites were unreachable, gated on a configuration value nothing ever set. The whole of Sources/SystemExtension/ was unreachable even from its own executable, its types shadowed by same-named stubs. usbipd status had 600 lines of reporting behind a guard that always took the early return.

None of it could have worked. OSSystemExtensionRequest resolves extensions inside the calling process's own bundle and requires that bundle to live in /Applications; a Homebrew-installed binary is neither. The claiming strategy underneath — setting IOMatchCategory and IOProbeScore from userspace — was measured not to unbind anything, and is not how a DriverKit extension matches in any case.

Nothing that worked has changed. Devices macOS has not bound a driver to are served exactly as before; devices whose interfaces it holds are refused by bind, which names the owner. Those decisions were always made by DeviceOwnershipInspector and by IOKit refusing USBInterfaceOpen, not by the extension.

User-visible changes:

• USBIPDSystemExtension.systemextension.tar.gz is no longer published. It was signed, notarized and stapled every release, and was not a System Extension bundle by structure — no NSExtension dict, no IOKitPersonalities, and an empty embedded.provisionprofile.
• systemextension_url and systemextension_sha256 are gone from the Homebrew metadata. Nothing consumed them.
• usbipd status reports how devices are accessed and which are shared, and no longer claims to report on an extension. --health is still accepted and says plainly that there is nothing further to check.
• bind and unbind no longer print "Checking System Extension status..." or branch their success message on whether a manager existed.
• ~/Library/Application Support/usbipd-mac/cli-claimed-devices.json is no longer read or written. It recorded claims that were never taken; any existing file is inert and can be deleted.
• usbipd.entitlements is gone. The binary carried com.apple.developer.system-extension.install, a restricted entitlement AMFI kills without an authorising profile, so release signing had to withhold it — which it always did. With nothing to install, signing has one path.

Sources/SystemExtension/Info.plist, its entitlements and the README explaining the failure are preserved under Documentation/development/system-extension-archive/. They record the identifier and entitlement decisions a future DriverKit extension needs.

Also corrected: CLAUDE.md claimed Tests/SharedUtilities/ was compiled by both test targets. SwiftPM ignores sources: paths outside the target directory, so it never was.

[v0.6.0] - 2026-08-09

This release makes usbipd usable on a machine with a USB hub, which in practice means most machines. It also makes CMSIS-DAP debug probes work end to end, and stops two attached devices from interfering with each other. Every fix below was measured against hardware: a SEGGER J-Link, a Raspberry Pi Debug Probe, an FTDI Quad RS232-HS, a Silicon Labs CP2102N, and a Pixel 10a.

Fixed — every device behind a hub was unreachable

A busid became a hub port path when locationID decoding was corrected, so a device two tiers down is 32-2.2. The wire protocol has no room for that: USB/IP identifies a device by a devid the client composes from the bus and device numbers in the import reply. The encoder kept only the final port, so 32-2.2 and 32-2 both went out as device number 2, and the server resolved the devid to whichever matched first.

That was the hub. Importing a CP2102N at 32-2.2 handed back an identifier naming the USB2.0 hub above it, and the first transfer died in USBDeviceOpen with kIOReturnExclusiveAccess, because macOS owns hubs. On a docked machine this was nearly every device.

The port path is now packed one nibble per tier, mirroring how macOS builds locationID, and a devid is resolved by matching it against the devices actually attached rather than by unpacking it — so a stale or unrepresentable identifier resolves to nothing and the client is told ENODEV instead of being pointed at the wrong device. Four tiers of up to fifteen ports each are representable; deeper chains are refused rather than approximated.

Fixed — transfers on one device blocked each other

Every transfer on an interface went through a single serial queue, and IOKit's ReadPipeTO and WritePipeTO block the calling thread until they complete or time out. A pending IN transfer therefore held the queue for its entire timeout, and no other transfer could start — including the OUT transfer carrying the command that would have made the device answer the read. Any protocol that reads and writes at once deadlocked until the timeout, then failed.

A Raspberry Pi Debug Probe showed it exactly: probe-rs posted a read, then fifteen writes that sat untouched for sixty seconds and all completed within sixteen milliseconds of the read finally timing out.

Each endpoint now has its own queue. Transfers on one endpoint stay ordered, as USB requires of a pipe, while unrelated endpoints proceed in parallel — which USB/IP has always assumed, since the protocol carries a sequence number precisely so replies may return out of order.

Fixed — a cancelled transfer was carried out anyway, and answered later

A client that withdrew a request got it performed regardless. The sequence number was claimed only after an IOKit enumeration, so an UNLINK arriving a millisecond later found nothing and was told the request had already completed; the abort was aimed at the endpoint in the UNLINK message, which is always zero, so it hit the control pipe; and AbortPipe was called with an endpoint number rather than an IOKit pipe reference. The transfer ran on, collected the answer to a later command, and delivered it under a sequence number the client had retired.

This is why probe-rs info failed against a CMSIS-DAP probe with "Error in the USB access": probe-rs drains the IN endpoint with a short timeout before its first command, and that abandoned read consumed every response it wanted. It matters well beyond debug probes, because libusb cancels any read it has timed out.

A cancelled request is now aborted at the correct pipe and answered by its RET_UNLINK alone, carrying -ECONNRESET as the URB status — which libusb reads as cancelled, and for a timed-out read as a timeout. probe-rs info against a Debug Probe now completes 946 transfers, 929 of them on the CMSIS-DAP bulk endpoints, and stops only at chip detection — the identical error the probe gives plugged straight into the Mac, because no target is wired to its SWD pins.

Fixed — two attached devices interfered with each other

USB/IP numbers requests per connection and the Linux client opens one connection per attached device, so every client starts again at 1. Request tracking used one process-wide table keyed on the bare sequence number, so a second client's first request looked like a duplicate of the first client's and was refused — and refused silently, because the rejection was thrown past the code that turns a refusal into a reply, leaving the client waiting for an answer that never came.

Attaching two devices from a single machine was enough to trigger it. Measured with an FTDI and a CP2102N driven at once: five of ten concurrent transfers failed before the change, none after.

Fixed — bind and unbind needed a daemon restart

They are separate processes that write a file the daemon read only at startup. A device bound while the daemon was running stayed unimportable — usbip attach answered "Request Failed" — and nothing said a restart was needed. The daemon now notices the file has changed and re-reads it.

Fixed — bind rewrote unrelated configuration

The bound-device list lived in the configuration file beside the port, the log level and the transfer tuning, so bind rewrote the whole file to append one busid. Because the CLI falls back to a default configuration when that file will not parse, a single malformed character turned the next bind into a silent reset of the port and log level.

The list now lives in ~/.usbipd/bound-devices.json and is the only thing bind and unbind touch; ~/.usbipd/usbipd-config.json is read and never written. An allowedDevices key left in an old configuration is migrated once and then ignored. Updates are serialised with a lock, so two bind commands can no longer lose each other's device.

Fixed — bind accepted devices it could not serve

A device that exposes no USB interfaces was reported as free, then failed every transfer with "Interface 0 not found". This happens when macOS has not configured the device, usually because another process is driving it directly — a browser tab using WebUSB, for instance. It is now refused with an explanation.

Fixed — USB-serial adapters were refused, and they work

bind decided ownership from which driver was attached to an interface. That inference holds for HID and mass storage, and is wrong for serial: IOUserSerial sits on every FTDI and CP210x interface without taking exclusive access. Measured against an FTDI Quad RS232-HS and a Silicon Labs CP2102N, both return their real descriptor over USB/IP and the FTDI returns real bulk data — status=0 actual=2 data=0260, its modem-status bytes.

So the most-requested category was being refused with a message claiming it needed a DriverKit entitlement from Apple. It does not.

Ownership is now decided by attempting to open the interface and closing it again, which is what the entitlement harness always did. The refinement "dexts do not block, kernel drivers do" is also wrong — AppleUserHIDDevice is a dext and does block — so nothing short of attempting the open is reliable. HID, mass storage and audio are still correctly refused.

Caveat: macOS keeps its serial driver attached and still publishes /dev/cu.*. Nothing prevents a local program opening that while a remote client drives the same UART.

Known limitations

Interrupt endpoints remain untested — no unbound interrupt device has been available. Isochronous is structurally incomplete: alternate settings are never selected and pipes are discovered once at open, so a UVC device's isochronous endpoints would never appear. Hub port paths deeper than four tiers cannot be represented in a devid and are refused. A device that disappears mid-session may keep failing until the daemon is restarted.

[v0.5.4] - 2026-08-08

Fixed — v0.5.3 could not serve any device

v0.5.3 stopped starting the System Extension manager at launch, which was correct: it cannot be activated from a shipping install and it was scanning the filesystem on every command. But three separate code paths still required a successful claim from it, so with the manager unstarted they all failed:

• bind refused every device with "System Extension failed to claim device"
• the import handler refused every attach with "Failed to claim exclusive access"
• every transfer failed with "Device not claimed for USB operations"

None of those claims did anything real. Devices are served through IOKit from userspace, and what actually gates access is the bind allow-list and IOKit's own exclusive-access rules. All three now proceed when no claim is recorded.

This was shipped, and would have been caught before tagging by running Scripts/verify-hardware.sh — which could not run, because no device was attached.

Fixed — composite debug probes were refused

Ownership was decided for the device as a whole, so any interface held by a driver refused the entire device. A Raspberry Pi Debug Probe pairs a free CMSIS-DAP interface with a CDC serial port that macOS claims, and was rejected despite the debug interface being usable — as are most composite probes, including ST-Link with VCP.

Ownership is now per interface. A device with at least one free interface can be bound, and bind says which interfaces are served and which macOS holds.

Known limitation — CMSIS-DAP probes enumerate but do not fully drive

probe-rs list finds a Raspberry Pi Debug Probe over USB/IP and reads its serial, and control transfers return its real descriptor. probe-rs info fails: the CMSIS-DAP exchange on endpoints 0x04/0x05 writes successfully but some reads return EPROTO or time out. Not yet diagnosed.

[v0.5.3] - 2026-08-08

Fixed — the CLI was unusable to read

A plain usbipd list printed roughly forty lines of debug logging around three lines of output. Every component constructs its own Logger with a hardcoded level — 27 at .info, 3 at .debug — so ServerConfig.logLevel governed nothing. Logger now has a process-wide floor defaulting to .warning, which wins where it is stricter. The daemon raises it to .info, since its output goes to a log file, and USBIPD_LOG_LEVEL overrides both.

Fixed — usbipd -v reported the wrong build

A Homebrew-installed release binary printed Build: Development, because main.swift emitted that literal. A correct implementation existed in CommandLineParser but was never reached, since main.swift intercepts -v first. Both now read one value.

Fixed — the CLI hunted for a System Extension on every command

The bundle detector walked the filesystem on every invocation, including the developer's own .build directory: an installed binary was scanning a source tree it found there and adopting a debug bundle from it. The manager was also started each time, restoring claim state for a subsystem that cannot be activated from a Homebrew prefix. Neither now happens, and usbipd status no longer instructs users to install a System Extension they cannot install.

Fixed — usbipd completion wrote files when given no verb

The action defaulted to generate, whose output directory defaults to the working directory, so a bare usbipd completion silently wrote three completion scripts wherever the user was standing. It now prints usage.

[v0.5.2] - 2026-08-07

Added — Intel Macs are supported again

Releases through v0.5.1 shipped an arm64-only binary, because the release workflow ran a plain swift build on an Apple silicon runner. Installing on an Intel Mac succeeded and then failed with Bad CPU type in executable (#32).

The build now produces both architectures and splits them with lipo, publishing usbipd-<version>-macos-arm64 and usbipd-<version>-macos-x86_64. The formula selects with on_arm/on_intel, so each user downloads one artifact of about 5.4 MB — slightly smaller than the arm64-only build, rather than the 10.8 MB a universal binary would cost everyone. Each slice is signed separately, since lipo -thin strips the signature.

[v0.5.1] - 2026-08-07

Fixed — control transfers were broken in release builds only

performControlTransfer returned a pointer out of setupPacket.withUnsafeBytes, which only guarantees it inside the closure. Every subsequent read was of freed memory. Debug builds happened to leave the bytes intact, so all prior validation passed; optimised builds did not, so bmRequestType, bRequest, wValue, wIndex and wLength became garbage, the device received a malformed request, and the pipe stalled — kIOUSBPipeStalled, reaching clients as EPROTO.

This affected v0.5.0 as shipped: control transfers are how a client enumerates a device, so attaching from Linux would fail against the released binary even though bulk transfers worked. Found by testing the Homebrew-installed artifact rather than a local debug build.

[v0.5.0] - 2026-08-06

First release in which the software does what it claims. Every prior release shipped a transfer path that had never moved data, and a bind that recorded devices without serving them. Those releases have been deleted rather than left to mislead.

The version number also means something now: it was hardcoded as "0.1.0" in three separate print statements, so every previous release reported 0.1.0 regardless of its tag.

Fixed — USB transfers now reach real hardware

The transfer path had never moved data. Several defects sat between a client request and the device, each hiding the next.

• ReadPipeTO's size argument is in/out — buffer capacity in, bytes read out — and was initialised to 0 and never set. IOKit was told the buffer could hold nothing, so any packet the device sent overran it. Every IN transfer on the bulk and interrupt paths failed this way.
• Transfer type was guessed from the request's interval field. USB/IP carries no transfer type, so it now comes from the device via GetPipeProperties. A J-Link declares bInterval 1 on both of its bulk pipes, so the guess routed bulk traffic to the interrupt path.
• IOKit addresses pipes by a 1-based index into the interface, not by endpoint number. Endpoint discovery is now implemented and the index looked up.
• Direction was taken from bit 7 of the endpoint address, which a conforming client leaves clear because it sends direction in its own field.
• USBSubmitProcessor fabricated a 0000:0000 device instead of resolving the devid, so every lookup failed with ENODEV.
• 33 escaped string interpolations in USBDeviceCommunicatorImplementation — the one in deviceIdentifier(for:) gave every device the same literal key, so no claim matched.

Verified with probe-rs driving a J-Link from a Linux client over USB/IP, reading the probe's VTref exactly as it does connected directly.

Fixed — the bind allow-list is enforced

bind wrote devices to allowedDevices and nothing ever read it back. The server advertised every USB device on the machine and served transfers on any of them, bound or not. isDeviceAllowed also returned true for an empty list, so a fresh install offered everything. Both are fixed; sharing is opt-in.

Added — bind refuses devices something else owns

Previously it allow-listed anything and failed later at transfer time. It now reads ownership from the IORegistry and refuses with an explanation naming the owner, distinguishing a kernel driver (needs a DriverKit entitlement) from a userspace holder (quit the app).

Added — configuration settings are honoured

maxUSBBufferSize, usbOperationTimeout, maxTotalConcurrentRequests, maxPendingURBsPerDevice, maxConnections and connectionTimeout all persisted and validated their own ranges while nothing read them. They now apply. maxUSBBufferSize matters most: IOKitUSBInterface allocated Int(bufferLength) straight from a wire-supplied UInt32. autoBindDevices was removed rather than implemented, being the opposite of the opt-in model.

Fixed — transfers reported bytes they did not have

A timed-out IN transfer answered with actualLength set to the buffer capacity and no payload, because that field is primed for ReadPipeTO's in/out contract and IOKit does not reset it when nothing is read. A client reading that many bytes would desynchronise. Zeroed on every non-success path, for bulk and interrupt.

The default transfer timeout was also raised from 5000 ms to 60000 ms. At five seconds any merely idle IN endpoint produced ETIMEDOUT, and adb gave up on its first read. Both were found by testing against an Android device.

Fixed — miscellaneous

• Bulk transfers apply a timeout; they used ReadPipe/WritePipe, which block forever
• kIOUSBTransactionTimeout maps to ETIMEDOUT rather than falling through to EPROTO
• A defer in a registry loop released the entry the next iteration was about to read
• CommandHandlerError was re-wrapped in itself, doubling error prefixes
• Data races in USBSubmitProcessor (concurrent-queue writes without .barrier)

Removed

• attach and detach, which macOS cannot implement — that is a client-side operation
• install-system-extension, which could not succeed from a Homebrew prefix, and diagnose, which did not complete in 300 seconds. Both were registered but absent from usbipd --help
• The .kiro and .gemini directories
• URBTracker, DeviceMonitor, the global logging functions, and five other mechanisms that had tests but no production callers, dropping 42 tests that covered code the daemon never reached
• Two message validators with no callers anywhere, one of which contained a check that could never fire (endpoint & 0x0F > 15)
• ~17,800 lines of test targets that no target compiled and that had never run

Added — MIT LICENSE

The README had claimed MIT while no LICENSE file existed, leaving the terms of distribution unstated.

Measured

• Seizing an interface does not take it from a kernel driver: USBInterfaceOpenSeize returns the same kIOReturnExclusiveAccess as a plain open. Neither unmounting nor ejecting releases a device either. Measured against mass storage, HID and a webcam.
• com.apple.security.device.usb is an App Sandbox entitlement and grants nothing here; the DriverKit transport entitlements are what gate this.

Known limitations

• Clients whose protocol keys off connection or reset events may not work, even on a device that claims cleanly. adb reports an otherwise-working Pixel offline: the phone announces itself once per USB connection and macOS already received that announcement, and attaching from a client causes no bus reset the phone can observe. Request/response devices are unaffected.

• UNLINK does not abort an in-flight read. It answers RET_UNLINK immediately, but the transfer is only reclaimed when its own timeout expires. This is why the timeout cannot be removed in favour of unlimited waits.

• bind does not affect a running daemon, which reads the allow-list only at startup.

• Interrupt endpoints are unverified against hardware

• Isochronous is structurally incomplete: alternate settings are never selected and pipes are discovered once at open, so a UVC device's isochronous endpoints never appear

• The System Extension subsystem is quarantined and no shipping path activates it

[v0.1.34] - 2025-08-28

Added

• Comprehensive shell completion system for bash, zsh, and fish
• Dynamic completion for device IDs, IP addresses, and ports
• Intelligent command-aware argument completion
• Homebrew automatic completion installation and configuration

Improved

• Enhanced CLI usability with tab completion support
• Better development experience with shell integration

[v0.1.26] - 2025-08-22

Added

• System Extension Bundle Distribution: Release workflow now builds and distributes system extension bundles
  • Builds USBIPDSystemExtension target explicitly in release workflow
  • Creates proper system extension bundle structure with Info.plist
  • Applies code signing to system extension bundle
  • Includes system extension bundle in release artifacts as compressed tar.gz
  • Generates checksums for both CLI binary and system extension
  • Updates Homebrew metadata to include system extension download URLs and checksums
  • Enhances release notes with system extension installation instructions

Fixed

• Homebrew System Extension Support: Fixed "System Extension Status: Not Available" error for Homebrew users
  • System extension bundle now properly distributed with releases
  • Homebrew formula can install complete functionality including system extensions
  • Users can now successfully run sudo usbipd install-system-extension

[v0.1.19] - 2025-08-21

Fixed

• Test Compilation: Fixed duplicate BundleSearchResult struct causing compilation ambiguity
  • Removed duplicate struct definition in test file
  • Now uses internal struct from main source file

[v0.0.17] - 2025-08-21

Fixed

• Release Automation: Complete resolution of GitHub Actions workflow automation issues
  • Fixed critical YAML syntax errors preventing workflow execution
  • Resolved conditional logic in job dependencies
  • Fixed GitHub Actions context variable usage in shell environments
  • Restored proper tag-triggered releases vs branch push filtering
  • Enabled manual workflow_dispatch testing with parameter validation

Improved

• Enhanced error handling in release workflows
• Improved homebrew-releaser integration reliability
• Better validation of release triggers and conditions

[v0.0.14] - 2025-08-21

Added

• Homebrew Releaser Migration: Complete migration from webhook system to homebrew-releaser GitHub Action
  • Automated homebrew tap management with homebrew-releaser action
  • Comprehensive post-migration monitoring and validation
  • End-to-end release automation with artifact management
  • Rollback procedures and validation scripts
  • Enhanced documentation and troubleshooting guides

Changed

• Updated release workflows to use homebrew-releaser action
• Improved homebrew tap repository management
• Enhanced release monitoring and validation

Removed

• Legacy webhook system for homebrew tap management

[v0.0.6] - 2025-08-18

Added

• Complete System Extension Installation Framework: Comprehensive system extension lifecycle management
  • Advanced installation orchestration with automatic and manual installation modes
  • Intelligent installation verification and validation system
  • Service lifecycle management with recovery and fallback mechanisms
  • Enhanced diagnostic capabilities for troubleshooting installation issues
• Production-Ready Installation Workflows: End-to-end system extension deployment
  • Automated installation process for developer environments
  • Manual installation guidance with step-by-step user instructions
  • Cross-platform compatibility validation (Intel, Apple Silicon, Universal)
  • Installation status monitoring with real-time progress reporting
• Enhanced CLI Commands and Diagnostics: Comprehensive command-line interface
  • New install-system-extension command for manual installation
  • Advanced diagnose command with system health checks
  • Installation verification and status reporting capabilities
  • Interactive troubleshooting and guidance system

Fixed

• Critical USB/IP Protocol Issues: Resolved fatal crashes and protocol errors
  • Fixed USB/IP device list request command encoding (0x05 → 0x8005)
  • Resolved "Index out of range" crashes in integration tests
  • Improved timing for asynchronous notification processing
  • Enhanced error handling for network communication failures
• Test Infrastructure Stability: Comprehensive test suite improvements
  • Fixed critical integration test failures affecting production validation
  • Improved mock device simulation and timing reliability
  • Enhanced test environment setup and teardown procedures
  • Resolved Swift concurrency compliance issues
• System Extension Bundle Detection: Enhanced bundle management
  • Improved bundle detection and validation logic
  • Fixed bundle path resolution in various installation scenarios
  • Enhanced error handling for bundle creation and management
  • Better compatibility with different macOS versions and architectures

Changed

• Installation Architecture: Streamlined and consolidated installation process
  • Simplified installation workflow with fewer user interaction points
  • Improved error recovery and fallback mechanisms
  • Enhanced logging and diagnostic information collection
  • Better integration with macOS System Extension APIs
• Test Infrastructure: Modernized testing approach
  • Consolidated test environments (development, CI, production)
  • Improved QEMU integration testing infrastructure
  • Enhanced mock systems for reliable automated testing
  • Streamlined CI/CD pipeline for faster validation

Security

• Enhanced System Extension Security: Improved security validation
  • Strengthened code signing and entitlement validation
  • Enhanced installation verification to prevent tampering
  • Improved system permission management and validation
  • Better isolation and sandboxing of system extension components

Performance

• Optimized Installation Process: Faster and more reliable installation
  • Reduced installation time through parallel processing
  • Improved resource utilization during installation
  • Enhanced memory management and cleanup procedures
  • Better error recovery without requiring system restarts

[v0.0.4] - 2025-08-18

Added

• Homebrew System Extension Integration: Complete end-to-end System Extension support for Homebrew installations
  • Automatic System Extension bundle creation during brew install process
  • Intelligent installation automation with developer mode detection
  • Manual installation fallback with guided user instructions
  • Cross-platform compatibility validation (Intel x86_64, Apple Silicon ARM64, Universal)
  • macOS version compatibility checking (macOS 11.0+)
• Enhanced Installation Automation
  • Automatic System Extension installation in developer mode environments
  • Manual installation script generation for standard environments
  • User guidance and troubleshooting documentation
  • Installation status monitoring and progress reporting
• Comprehensive Error Handling and Recovery
  • Structured error categorization and handling framework
  • Automatic recovery strategies for common installation failures
  • Diagnostic information collection for troubleshooting
  • Environment-specific validation and setup verification
• Advanced Testing Infrastructure
  • Complete end-to-end Homebrew workflow testing
  • QEMU integration testing for USB/IP protocol validation
  • Multi-environment test suite (development, CI, production)
  • Comprehensive compatibility testing across architectures and macOS versions

Fixed

• CI/CD Pipeline Reliability
  • Resolved GitHub Actions workflow parameter mismatches
  • Fixed SwiftLint strict mode violations (31 violations resolved)
  • Corrected test discovery and execution in CI environments
  • Improved code quality with comprehensive linting compliance
• System Extension Bundle Creation
  • Fixed bundle structure validation and code signing integration
  • Resolved duplicate type definitions and build conflicts
  • Corrected Info.plist generation and entitlements configuration
• Build System Improvements
  • Enhanced Swift Package Manager integration
  • Improved dependency resolution and caching
  • Fixed cross-compilation issues for Universal binaries

Changed

• Test Architecture Refactoring
  • Reorganized test helper functions to prevent XCTest discovery conflicts
  • Improved test naming conventions (test* → validate* for helpers)
  • Enhanced test execution performance and reliability
• Documentation and Code Quality
  • Comprehensive spec-driven development documentation
  • Improved code commenting and inline documentation
  • Enhanced error messages and user-facing guidance

Technical Details

• System Extension bundle creation with proper macOS bundle structure
• CFBundlePackageType correctly set to "SYSX" for System Extension identification
• Automated code signing integration for notarization-ready bundles
• Environment detection for development vs. production installation workflows
• QEMU-based protocol validation for comprehensive testing coverage

[v0.0.3] - 2025-08-16

Added

• Uploaded certs for code signing

[v0.0.2] - 2025-08-16

Added

• Improved System Extension installation process

[v0.0.1] - 2025-08-16

Added

• Initial release