Skip to content

Changelog

For the complete, detailed changelog, see CHANGELOG.md in the repository root.

Recent Releases

[Unreleased]

Man Page Integration & Production Readiness (v0.2.0)

Added:

  • Man Page Integration: New /cmd command that reads system man pages and generates exact commands from natural language queries
  • Intelligent command detection (60+ common CLI tools)
  • Automatic man page parsing for accurate command generation
  • Fallback to general CLI knowledge when man pages unavailable
  • Supports find, grep, sed, awk, tar, curl, git, docker, kubectl, and more
  • Interactive Setup Wizard: Beautiful guided first-run experience
  • Four model presets: Fast (0.5B), Balanced (1.5B), Best (3B), Premium (7B)
  • Clean single-line progress bar (no more 500-line spam)
  • Production-grade downloads with retry logic and resume support
  • Disk space validation before download
  • Optional post-setup quick test
  • Enhanced Model Selection: Switched default from qwen3-4b (2.6GB) to qwen2.5-1.5b (1.0GB)
  • 22% faster inference (5-8s avg response vs 6-10s)
  • 61% smaller download size
  • Improved quality with better tool calling support
  • Built-in Review Command: Code review functionality now built into core
  • Multi-Platform Distribution: Automated releases via GoReleaser for macOS, Linux, and Windows
  • Homebrew Support: Official Homebrew tap at scmd/tap
  • npm Distribution: Cross-platform installation via scmd-cli npm package
  • Linux Packages: Native deb, rpm, and apk packages
  • Install Scripts: Universal wget/curl installation script with checksum verification
  • Docker Images: Multi-arch Docker images on Docker Hub
  • Shell Completions: Auto-generated completions for bash, zsh, fish, and PowerShell
  • Release Automation: GitHub Actions workflow for automated releases
  • Comprehensive Documentation:
  • Detailed installation guide (INSTALL.md)
  • Release process documentation
  • Distribution infrastructure guide

Installation Methods:

# Homebrew (macOS/Linux)
brew install scmd/tap/scmd

# npm (Cross-platform)
npm install -g scmd-cli

# Shell script (wget/curl)
curl -fsSL https://scmd.sh/install.sh | bash

# Linux packages
# Debian/Ubuntu
sudo dpkg -i scmd_VERSION_linux_amd64.deb

# Red Hat/Fedora/CentOS
sudo rpm -i scmd_VERSION_linux_amd64.rpm

Enhanced:

  • Model Performance: Optimized llama.cpp configuration for 22% faster inference
  • Context size increased from 1024 to 8192 tokens
  • Flash attention enabled
  • Continuous batching for multiple requests
  • Memory locking for consistent performance
  • Optimized KV cache (F16)
  • Model Downloads: Production-grade download system
  • Retry logic with exponential backoff (3 attempts)
  • Resume support using HTTP Range headers
  • Disk space validation (1.2x file size buffer)
  • Enhanced error messages with structured help
  • 7B Model URL: Fixed qwen2.5-7b download (changed from Q4_K_M to Q3_K_M)
  • Single file download (was multi-part)
  • Reduced size from 4.7GB to 3.8GB
  • Updated README with all installation methods, benchmarks, and new features
  • Makefile with release and distribution targets
  • Shell completion generation command

Fixed:

  • Setup wizard download URLs (now use DefaultModels instead of hardcoded URLs)
  • 7B model multi-part download issue
  • Context size errors on large files
  • Progress bar display (clean single-line instead of spam)

Performance Benchmarks (M1 Mac, 8GB RAM):

Model Avg Response Tokens/sec Quality
qwen2.5-0.5b 3-5s ~45 tok/s ⭐⭐⭐
qwen2.5-1.5b 5-8s ~30 tok/s ⭐⭐⭐⭐
qwen2.5-3b 8-12s ~18 tok/s ⭐⭐⭐⭐⭐
qwen2.5-7b 15-25s ~8 tok/s ⭐⭐⭐⭐⭐

[0.1.0] - 2025-01-06

Initial release with core functionality.

Added:

  • Offline-first AI-powered slash commands
  • llama.cpp integration with Qwen models
  • Auto-download of models on first use
  • Built-in /explain command
  • Repository system for command distribution
  • Shell integration for bash, zsh, and fish
  • Multi-backend support (llama.cpp, Ollama, OpenAI, Together.ai, Groq)
  • Command composition and chaining
  • Configuration management
  • Model management (list, pull, remove, info)
  • Slash command system
  • Command lockfiles for reproducibility
  • Context gathering and caching

Security:

  • Input validation and sanitization
  • Path traversal prevention
  • Resource limits for LLM inference
  • Secure model downloads with checksum verification
  • Comprehensive security documentation

Documentation:

  • README with quick start guide
  • Architecture documentation
  • Security documentation
  • Troubleshooting guide
  • API documentation
  • Contributing guidelines

Version History

For detailed version information and complete release notes, see:

Release Schedule

scmd follows semantic versioning and releases on an as-needed basis:

  • Major releases (v2.0.0): Breaking changes, significant new features
  • Minor releases (v1.1.0): New features, backward compatible
  • Patch releases (v1.0.1): Bug fixes, backward compatible

Upgrade Guide

Upgrading to Latest Version

brew upgrade scmd

npm update -g scmd-cli

curl -fsSL https://scmd.sh/install.sh | bash

# Debian/Ubuntu
sudo apt upgrade scmd

# Fedora/RHEL
sudo dnf upgrade scmd

cd scmd
git pull origin main
make build
sudo make install

Staying Updated

  • Watch Releases: GitHub Watch → Releases only
  • Star the Repo: Get updates in your GitHub feed
  • Follow Changelog: Check this page regularly for updates

Deprecation Policy

When we deprecate features:

  1. Advance notice: Minimum 1 major version before removal
  2. Migration guide: Clear documentation on alternatives
  3. Warnings: Deprecation warnings in the application
  4. Support period: Bug fixes for deprecated features during transition

Security Updates

Security updates are released as patch versions as soon as fixes are available. For security advisories, see:

Contributing to the Changelog

When contributing, please update CHANGELOG.md following the Keep a Changelog format:

## [Unreleased]

### Added
- New feature X

### Changed
- Modified Y for better performance

### Fixed
- Bug in Z

### Security
- Patched vulnerability in A

See Contributing Guidelines for more details.