v0.1.10

v0.1.10

✨ Features

• ACP Support Documentation
  • Added ACP (Agent Client Protocol) support documentation to READMEs
  • VibeCoding can run as an ACP stdio agent for editor integrations
  • Compatible with VS Code, Zed, and JetBrains IDEs (IntelliJ IDEA/WebStorm) via ACP plugins

📖 Documentation

• Updated main README.md with ACP support feature
• Updated English README with features section
• Updated Chinese README with features section

---

v0.1.8

v0.1.8

🐛 Bug Fixes

• Token Counting with Cache-Aware TotalTokens
  • Fixed Anthropic TotalTokens calculation to include CacheRead and CacheWrite tokens
  • Added PromptTokens() and TotalInputTokens() helper methods to Usage struct
  • Updated CacheInfo() to use TotalInputTokens() as denominator for accurate cache hit rates
  • Updated TUI to display correct token counts including cache tokens

🧪 Testing

• Added comprehensive tests for PromptTokens() and TotalInputTokens() helper methods
• Updated Anthropic provider tests with TotalTokens validation

---

v0.1.6

v0.1.6

✨ Features

• Session Management Command
  • Added /sessions command for browsing and managing project sessions
  • Supports listing, switching, clearing, and deleting sessions
  • Shows session details including file path and message count

🐛 Bug Fixes

• Sandbox Initialization
  • Fixed sandbox initialization validation and bwrap multiarch compatibility
  • Improved error handling for sandbox setup

📖 Documentation

• Updated AGENTS.md with current version information
• Formatted Go code for consistency

v0.1.5

v0.1.5

✨ Features

• DeepSeek V4 Default Models
  • Updated default model specs to DeepSeek V4 (Flash and Pro)
  • 1M context window, up to 384K max output tokens
• Install Script Improvements
  • Install scripts now show config directory path on completion

🐛 Bug Fixes

• Windows IME Support
  • Fixed Windows IME (CJK input) support in terminal
  • Fixed shell command resolution on Windows
  • Added config loading diagnostics for troubleshooting
• Musl Deb Packages
  • Fixed invalid dpkg architecture names for musl deb packages

🛠 Improvements

• Configuration Simplification
  • Removed auth.json support — all credentials now in settings.json only
  • Cleaner config path with single source of truth

📖 Documentation

• Clarified that OpenAI/Anthropic API-compatible services are also supported
• Removed all auth.json references from docs and install scripts
• Added expanded Windows %APPDATA% path examples
• Clearly distinguished Windows vs Linux/macOS config paths

---

v0.1.4

v0.1.4

✨ Features

• Linux musl Build Support
  • Added make build-linux-musl target for statically linked musl binaries (amd64 + aarch64)
  • musl tarballs produced via dist-tarball and dist targets
  • musl Debian packages produced via dist-deb target (amd64-musl / arm64-musl)
  • npm packages: vibecoding-installer-linux-musl-x64 and vibecoding-installer-linux-musl-arm64
  • npm uses libc field for proper musl/glibc resolution (npm >=9.4)
  • postinstall.js auto-detects musl vs glibc on Linux

---

v0.1.2

v0.1.2

✨ Features

• Prompt Cache Optimization

  • Implemented prompt cache optimization following LLM_Agent_Cache.md strategy
  • Cache system prompts and static context across multiple turns
  • Reduces API costs by reusing cached tokens for repeated prefixes

• TUI Markdown Syntax Highlighting

  • Assistant messages in TUI now have markdown syntax highlighting
  • Code blocks, headers, and formatting are visually distinguished
  • Improves readability of LLM responses

🐛 Bug Fixes

• Security & Correctness

  • Resolved critical security, race condition, and correctness issues
  • Addressed high and medium severity correctness issues across codebase
  • Removed dead code and improved overall code correctness

• TUI Stability

  • Fixed TUI startup hang caused by clearStdin blocking on unsupported stdin
  • Fixed TUI assistant message rendering broken by ANSI escape codes in prefix check

🛠 Improvements

• Code Quality
  • Addressed remaining medium severity issues across codebase
  • npm package versions updated

v0.1.1

Added

• Cache hit rate display: Footer shows cumulative cache hit percentage (highlighted when ≥50%)
• Per-turn cache info: Token usage line now includes cache read/write counts per turn
• Usage.CacheInfo(): Extracted shared cache display logic to eliminate 3× duplication
• Proxy compatibility: Handle proxies that send usage fields in message_delta instead of message_start
• OpenAI multi-chunk usage: Handle OpenAI proxies that split usage across multiple SSE chunks (first-wins per field)
• Print mode fix: Fixed missing space before $ in print-mode token summary line
• Cache tests: 37 unit tests for CacheInfo(), formatCachePercent(), and renderFooter() cache section
• Integration tests: 12 httptest integration tests for Anthropic and OpenAI SSE cache token parsing

Changed

• npm package versions now use v-prefixed format (e.g. v0.1.1)
• Normalized JSON formatting across all npm package.json files

v0.0.9

v0.0.9

✨ Features

• Image Support in Tools

  • read tool now supports reading image files (PNG, JPEG, GIF, WebP)
  • Images are returned as base64-encoded data with MIME type information
  • LLMs can now analyze and understand image content
  • Supported formats: .png, .jpg, .jpeg, .gif, .webp

• Rich Content Tool Results

  • New ToolResult struct supports both plain text and rich content blocks
  • Tools can now return text + images in a single result
  • New factory functions: NewTextToolResult() and NewImageToolResult()

• Model Switching

  • /model <id> command allows switching models in interactive mode
  • /model without arguments shows current model and available options
  • Agent resets automatically when model is switched

• Enhanced Help System

  • /help command now shows detailed command descriptions
  • Added keyboard shortcuts reference (Tab, Esc, Ctrl+O, PgUp/PgDn)

🛠 Improvements

• Context Token Estimation

  • Fixed double-counting issue when both Content and Contents are present
  • Image tokens estimated as ~1200 tokens per image

• Provider Message Conversion

  • OpenAI: Images in tool results sent as supplementary user messages
  • Anthropic: Images sent as separate user messages alongside tool_result

🧪 Testing

• Added TestReadToolImage test case for image reading functionality
• All tool tests updated for new ToolResult return type

v0.0.8

v0.0.8

✨ Features

• NPM Multi-Architecture Split Packages
  • Split the npm package from a single all-platform bundle (~60MB) into 6 platform-specific packages (~10MB each)
  • Users now only download the binary for their current platform, reducing install size by 83%
  • Uses npm optionalDependencies + os/cpu fields for automatic platform matching
  • Main package vibecoding-installer is only ~2KB, links the correct platform package via postinstall

🛠 Improvements

• Build System
  • Added scripts/build-npm-packages.sh to generate platform-specific npm packages
  • Added make npm-packages, make npm-pack, make npm-publish-all targets
  • sync-npm-version.sh now syncs versions across all platform packages

---

Full Changelog: v0.0.7...v0.0.8

v0.0.7

Changelog

v0.0.7

✨ Features

• Cross-Platform Sandbox Support

  • Sandbox now supports macOS and Windows in addition to Linux
  • macOS uses sandbox-exec for process isolation
  • Windows uses restricted process creation without network access
  • Platform-specific sandbox implementations selected automatically

• Repository Rename

  • Module path renamed to github.com/startvibecoding/vibecoding
  • All imports, documentation, and scripts updated accordingly

🛠 Improvements

• Platform-Specific Process Handling
  • Extracted SysProcAttr configuration into build-tagged files (bash_unix.go, bash_windows.go)
  • Background child process cleanup now works correctly on all platforms
  • Setpgid only set on Unix systems; Windows uses CREATE_NEW_PROCESS_GROUP

📖 Documentation

• Updated all GitHub URLs to new repository location
• Added v0.0.6 and v0.0.7 release notes

Full Changelog: v0.0.6...v0.0.7