@entro314labs/ai-changelog-generator

# Changelog All notable changes to this project will be documented in this file. The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/), and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html). ## [Unreleased] ### Added - **Vercel AI Gateway provider** (`vercel-gateway`): routes completions through the Vercel AI Gateway via the Vercel AI SDK, reaching every gateway-fronted model with one credential using `<provider>/<model>` ids. Authenticates with `AI_GATEWAY_API_KEY` or Vercel OIDC. Defaults to the pinned `deepseek/deepseek-v4-flash-0731` snapshot. All existing BYOK and local providers (OpenAI, Anthropic, Google, Azure, Bedrock, Vertex, Hugging Face, Ollama, LM Studio, GitHub Copilot) are unchanged. - New configuration keys: `AI_GATEWAY_API_KEY`, `AI_GATEWAY_BASE_URL`, `AI_GATEWAY_MODEL`, `VERCEL_OIDC_TOKEN`, and `AI_TEMPERATURE`. - `generateChangelog()` and `generateChangelogFromChanges()` now accept an options object, so embedders can control format, template, attribution, analysis mode and output path. - **Model response cache**: gateway responses are cached on the model boundary, keyed by a SHA-256 of model identity plus response-affecting call parameters, backed by an in-memory LRU over a JSON file cache under `~/.cache/ai-changelog-generator`. A `--dry-run` preview followed by a real run no longer pays for the same analysis twice. Disable with `AI_CACHE_ENABLED=false`. - **AI error classification**: provider failures are classified into actionable codes that record whether a retry could succeed and whether another model should be tried, so a rejected credential stops immediately instead of being retried like a rate limit. - Gateway responses with empty text are now reported as `output_truncated` / `no_output` failures instead of being returned as success. The default flash snapshot can finish with `length` and no visible text when the output budget is tight, which previously produced an empty changelog entry that looked like a successful generation. The gateway output budget was raised accordingly (a one-line summary was measured at 1120 tokens against the live gateway, so the previous 1000-token default truncated). ### Fixed - **Failed commits no longer damage staged work.** Rollback now snapshots the index with `git write-tree` and restores it with `git read-tree`. The previous path-based `git reset HEAD -- <paths>` unstaged hunks the user had staged before the run whenever a file was both partially staged and partially modified. - **AI opt-out is honored.** `--no-ai` / `includeAIAnalysis: false` reached the orchestrator but was dropped before commit processing, so providers were still called for default and tag-range generation, and in detailed/enterprise working-directory modes. - **Provider failures no longer delete commits.** A failed AI summary on the sequential path (<= 20 commits) discarded the commit entirely; only commits git cannot read are dropped now. This removes the asymmetry where a 21-commit release survived an outage a 20-commit one did not. - **Commit filters are applied.** `--since`, `--author` and tag ranges were accepted and then ignored by recent-commit analysis, which loaded unfiltered history and sliced it. - **MCP timeouts cancel work.** Tool calls now carry an `AbortController`, so a timed-out call can no longer write `AI_CHANGELOG.*` after the client was told it failed and retried. - **MCP reports write failures.** A requested file write that fails is now an error instead of a success, and writes are atomic (temp file + rename). - **MCP response matches the requested format.** JSON/HTML formatting was applied only to the file, leaving the returned content as Markdown. - **Strict commit validation works.** The returned validation result now describes the message actually returned (previously the pre-improvement one), and `strictValidation` blocks insertion instead of being ignored. - Removed unconditional `DEBUG` logging from changelog generation, and made progress output respect silent/MCP mode. ### Changed - **Migrated the MCP server to the v2 SDK** (`@modelcontextprotocol/server` 2.0.0, protocol revision 2026-07-28), replacing `@modelcontextprotocol/sdk` 1.x. Tools are now registered with `registerTool` and Zod schemas, so the SDK derives the advertised JSON Schema and validates arguments _before_ a handler runs. `generate_changelog` additionally declares an `outputSchema` and returns `structuredContent`, giving clients machine-readable results instead of prose. Cancellation now uses the protocol's own per-request `AbortSignal`, composed with the server-side time budget. - **Provider mixins no longer overwrite provider code.** `applyMixins` filled gaps by `Object.assign`-ing over the class prototype, which silently disabled every provider-specific method. Mixins are now applied only where a provider defines nothing. This activated correct behaviour that had never run: Ollama required a nonexistent `OLLAMA_API_KEY` (now `OLLAMA_HOST`) and its own tool-support check always reported false, so Ollama never sent tools; GitHub Copilot required `GITHUB-COPILOT_API_KEY` (now `GITHUB_COPILOT_TOKEN`); Bedrock required a `BEDROCK_API_KEY` rather than the AWS chain; and the `dummy`/`mock` development providers claimed to be GPT-5.2 with full capabilities. Roughly twenty stale duplicate model/capability definitions were removed so `model-config.ts` is the single source of truth. - **Refreshed every model default across all providers.** The registry still pointed at retired or near-retired models (`gpt-4.1`, `gpt-4o`, `o3`/`o4-mini`, dated `claude-*-4-5-2025xxxx` ids, `gemini-2.5`, `llama-3.3`). Current defaults: OpenAI GPT-5.6 (Sol/Terra/Luna), Claude 5 (`claude-opus-5` / `claude-sonnet-5` — undated ids, as dated variants are not valid), Gemini 3.x, Bedrock `anthropic.claude-*`, Qwen3 on Hugging Face, and DeepSeek V4 through the gateway. Model ids were verified against the live gateway catalogue (322 models) rather than assumed. Capability entries were added for the new families, since capability lookup is substring-based and the new ids matched nothing. Package keywords and the install banner no longer advertise retired models. - The Azure no-deployment-info path no longer overrides every tier with a hardcoded legacy deployment name; it falls back to the configured defaults. - The MCP server now reports its real package version; a fixed relative path resolved to a nonexistent file in the built layout, so it advertised `1.0.0` to every client. - Upgraded to TypeScript 7, `js-yaml` 5 (named exports; empty configs short-circuit to defaults), `openai` 7, `@google/genai` 2, `@anthropic-ai/sdk` 0.116, Vitest 4.1.10 and the current Oxc toolchain. - Removed 11 unused dependencies from the core package and 3 from the extension. - VS Code: the `model` setting now maps to `AI_MODEL` (it previously mapped to `OPENAI_MODEL` and so did nothing for non-OpenAI providers); `temperature` and the conventional-commit scope settings now reach the core library. - VS Code: release generation stops when there are no unreleased commits instead of proposing a patch bump. - Repaired the release and provider-test CI workflows, which could not succeed in their prior form. ### Security - VS Code webviews build result rows with `textContent` instead of `innerHTML`, so repository- controlled values (file paths, branch and tag names, validation text) cannot inject markup. ## [3.6.0] - 2025-11-27 ### Feat - **MCPB Support**: Added full support for Model Context Protocol Bundle (MCPB) extension format. - **Packaging**: Updated manifest and packaging for seamless MCP server integration. ### Fix - **Dry Run**: Fixed `dry-run` mode to correctly skip file writing while preserving output preview. - **CLI**: Fixed interactive mode timeouts in non-TTY/CI environments. - **MCP Server**: Fixed `ReferenceError` by using dynamic imports for `child_process` in ESM environment. - **Validation**: Fixed `WorkspaceChangelogService` validation to correctly detect git repositories. - **UI**: Fixed `EnhancedConsole.metrics` method shadowing in CLI UI utilities. - **Tests**: Fixed integration test failures in styling and workspace services. ### Test - **Coverage**: Validated full test suite with 696 passing tests. - **Mocking**: Fixed mocking strategies in styling integration tests. ### Chore - **Dependencies**: Updated project dependencies and type definitions. - **Audit**: Verified project robustness and MCP tool integration. ## [3.3.0] - 2025-10-19 ### Refactor - **Architecture**: Major refactor of git management with new `GitManager` and `CommitTagger` domains. - **Services**: Re-architected `AnalysisEngine`, `ChangelogService`, and `WorkspaceChangelogService` for better separation of concerns. - **Config**: Modernized `biome.json` and `package.json` configurations. ### Feat - **Git Analysis**: Enhanced merge commit analysis and processing. - **Configuration**: Added support for `merge` commit type in configuration. ### Fix - **Reliability**: Resolved intermittent test failures and improved test suite reliability. - **Error Handling**: Added comprehensive error classes and improved exception handling. ### Test - **New Tests**: Added `missing-functionality.test.js` covering edge cases and new architectural components. ## [3.2.1] - 2025-08-10 ### feat - add comprehensive test coverage with 100% test success rate - add missing method implementations across all service classes - add modern CI/CD pipeline with automated testing and publishing - add Biome linting configuration for code quality - add comprehensive error handling and validation - add robust CLI timeout handling and graceful failures - add professional documentation and provider setup guides ### fix - fix AI provider initialization and isAvailable method calls - fix configuration manager null safety with proper error handling - fix CLI test timeouts and hanging commands - fix missing boolean getters (hasAI, gitExists) in main facade - fix homepage URL typo in package.json - fix CI/CD workflow paths to match current project structure - fix Node.js version consistency across all configurations - fix Vitest reporter deprecation warnings ### test - add 658 comprehensive tests with 100% pass rate - add missing service method implementations for test compatibility - add resilient CLI end-to-end testing with proper timeouts - add comprehensive architecture and functionality testing - add provider integration testing and validation - add performance and memory usage testing - add cross-platform compatibility testing ### build - update to Node.js 22.x across all environments - add pnpm workspace configuration for monorepo structure - add Vitest testing framework with coverage reporting - add modern GitHub Actions workflows for CI/CD - standardize package management with pnpm ### docs - update comprehensive README with installation and usage - add provider setup and troubleshooting documentation - add environment variables and configuration guides - update contributing guidelines and development setup ### chore - update dependencies including version increments from 3.0.3 to 3.1.1 - downgrade @types/node from 24.2.0 to 22.17.1 for compatibility - remove demo media files (GIFs, MP4s) and test scripts - clean up unused variables in utils.js ## [3.1.2] - 2025-08-06 ### build - update package.json configuration ### ci - improve deployment configuration ## [3.1.1] - 2025-08-06 ### fix - fix release workflow version handling - update package configuration for proper publishing - add publishConfig for public npm access - remove references to non-existent test scripts ### refactor - update utility imports from 'consolidated-utils.js' to 'utils.js' - restructure MCP server entry point - remove sample DXT extension files ### docs - add comprehensive AI Provider Setup Guide - consolidate provider configuration instructions ### chore - add demo GIF files and video content - expand color constants with Dracula theme variants ## [3.1.0] - 2025-08-01 ### feat - add interactive commit workflow with AI-powered commit message generation - add commit staging and unstaging functionality - add branch intelligence analysis for commit type suggestions - add interactive file selection for staging - add conventional commit format validation - add advanced git repository analysis and health checks - add `ai-changelog commit` command with interactive staging - add dry-run mode and custom message support - add provider-specific commit message optimization - add real-time branch analysis and suggestions ### refactor - improve orchestrator for commit workflow support - extend application service with commit workflow methods - add branch intelligence and staging utilities to utils - add comprehensive commit command support to CLI controller ### perf - optimize git operations with better error handling ### feat - add interactive staging service for git operations - add commit message validation service ## [3.0.3] - 2025-08-06 ### fix - fix intermittent "No working directory changes detected" issue - resolve Azure OpenAI provider token counting issues - improve error handling and fallback logic throughout system - improve CLI output formatting with professional visual styling ### docs - add comprehensive JSDoc documentation ### perf - add robust handling of git edge cases and command failures - clean error messages without technical implementation details - add session management with deduplicated warnings - improve fallback behavior when operations fail ## [3.0.0] - 2025-07-31 ### feat - add support for 10+ AI providers (OpenAI, Anthropic Claude, Google, Azure OpenAI, Amazon Bedrock, Vertex AI, Ollama, LM Studio, Hugging Face) - add advanced code diff analysis and semantic understanding - add intelligent commit categorization and breaking change detection - add professional changelog formatting with multiple output formats - add Model Context Protocol (MCP) server for Claude integration - add real-time changelog generation through MCP tools - add secure API key management and provider switching - add interactive and non-interactive CLI modes - add multiple analysis modes (standard, detailed, enterprise) - add flexible output formats (Markdown, JSON, plain text) - add advanced git integration with branch analysis - add automatic model selection based on commit complexity - add provider failover and load balancing - add support for both cloud and local AI models - add configurable model parameters and optimization ### refactor - implement domain-driven design with clear separation of concerns - add provider abstraction with base provider interface - add model recommendation engine - add advanced git integration with repository health assessment ### docs - add extensive documentation and provider setup guides - add troubleshooting documentation and environment variable reference - add contributing guidelines with architecture overview ### build - add shell wrapper scripts for cross-platform compatibility - add automatic dependency detection ### feat - add YAML-based configuration system (ai-changelog.config.yaml) - add environment variable support with .env.local - add customizable output templates and formatting ## [2.5.1] - 2025-08-01 ### fix - clean git error messages to display user-friendly warnings - suppress stderr output from git commands - simplify model availability error messages with session-based deduplication - add `execGitShow()` method for safer operations with missing files - add graceful handling of deleted/renamed files ### feat - add `--health` CLI flag for repository health assessment - add `assess_repository_health` MCP tool for AI assistants - add 8-point commit message quality scoring system - add repository hygiene recommendations and actionable insights - add health scoring (0-100) with categorization ### perf - clean console output with reduced error spam - improve fallback messaging for model selection - add informative warnings with actionable suggestions - improve rule-based analysis reliability without AI providers ## [2.5.0] - 2025-08-01 ### feat - add repository health assessment with `--health` CLI flag and `assess_repository_health` MCP tool - add real-time commit message quality detection and scoring (8-point scale) - add automatic warnings for poor commit messages with actionable suggestions - add repository hygiene metrics ### fix - add graceful handling of missing files in git history - improve error recovery for git command failures - improve fallback analysis when git operations fail ### feat - improve rule-based analysis to work without AI providers - add commit message quality warnings to changelog output - add `assess_repository_health` tool with JSON output format - add color-coded health scores and proactive suggestions ## [2.4.0] - 2025-08-01 ### feat - add `generate_changelog_from_changes` MCP tool for working directory changelog generation - add support for all analysis modes and model overrides in MCP tools - add file writing behavior - MCP tools now write `AI_CHANGELOG.md` files to project root - add `includeAttribution` parameter to `analyze_current_changes` MCP tool schema ### fix - add attribution footer by default in MCP tools, matching CLI behavior - eliminate functionality differences between CLI and MCP interfaces ### feat - add `generateChangelogFromChanges()` and supporting helper methods - add `generateWorkingDirChangelog()` for AI-powered changelog generation - add `generateBasicChangelogFromChanges()` for rule-based fallback - add changelog file writing to MCP tool workflows ## [2.3.1] - 2025-08-01 ### feat - add optional attribution footer to generated changelogs - add `--no-attribution` CLI flag to disable attribution footer - add `includeAttribution: false` MCP parameter to disable attribution ## [2.3.0] - 2025-08-01 ### feat - add `--model` CLI flag and MCP parameter for forcing specific AI models - add model capability validation and feature support display - add graceful fallback with capability warnings - add `--comprehensive` CLI command and `analyze_comprehensive` MCP tool - add repository statistics: commit count, contributor analysis, age calculation - add branch analysis: local/remote branch enumeration, current branch detection - add working directory status: staged/unstaged/untracked file counts - add AI-powered health assessment with actionable recommendations - add `--branches` CLI command and `analyze_branches` MCP tool - add branch enumeration and unmerged commit identification - add dangling commit detection and cleanup recommendations - add `--untracked` CLI command with file categorization - add risk assessment and automatic .gitignore recommendations ### fix - fix multiline commit message handling in `parseCommitOutput()` - fix commit retrieval format parameter in CLI argument processing - add `analyzeRepositoryWithAI()` method implementation - correct `generateText()` to `generateCompletion()` API calls ### refactor - improve `selectOptimalModel()` to respect model override - update CLI help documentation with new commands - extend MCP server schema for new functionality ## [2.2.0] - 2025-08-01 ### feat - add Model Context Protocol server with 8 tools - add `generate_changelog` for AI-powered changelog generation - add `analyze_commits` for commit pattern analysis - add `analyze_current_changes` for staged/unstaged file analysis - add `get_git_info` for repository metadata and statistics - add `configure_ai_provider` for AI provider testing and validation - add `validate_models` for model availability and capability checking - add dual CLI/MCP architecture with shared core libraries - add intelligent model selection based on complexity with cost optimization - add TypeScript definitions in `types/index.d.ts` ### refactor - migrate from flat to modular package structure - update Azure OpenAI integration to v1 API with 2025-04-01-preview version - scope package to `@entro314-labs/ai-changelog-generator` ### build - add `@modelcontextprotocol/sdk@1.12.1` dependency ## [2.1.0] - 2025-08-01 ### feat - add interactive CLI with commit selection - add checkbox interface for commit cherry-picking - add staged/unstaged change analysis - add AI-powered commit message validation - add configurable analysis depth modes - add `--detailed` for comprehensive business and technical analysis - add `--enterprise` for stakeholder-ready documentation - add `--analyze` for current working directory analysis - add extended git operation support - add diff analysis with before/after context - add file categorization by type - add conventional commit parsing and validation - add rule-based analysis fallback when AI unavailable ### refactor - expand CLI options with version-specific changelog generation - add date range filtering and output format selection - extend AI model support with GPT-4 series and Azure OpenAI integration ### fix - improve conventional commit detection - improve handling of binary files - improve resilience to git command failures ## [2.0.0] - 2025-08-01 ### feat - add OpenAI and Azure OpenAI API support - add automated commit analysis and categorization - add configurable changelog output formats - add environment variable and config file support ### refactor - migrate from manual to automated changelog generation - redesign API for programmatic usage ### feat - add direct git command integration - add asynchronous batch processing - add template-based markdown generation - add comprehensive error recovery mechanisms ## [1.0.0] - 2025-07-31 ### feat - add basic changelog generation with git integration - add command-line argument processing - add markdown changelog generation - add npm publishing setup with public access ### fix - add publishConfig for npm publishing - remove missing script references in package.json - correct build and deployment processes ### feat - add git commit history parsing - add changelog file management - add CLI argument handling --- ## VDK CLI Component ### [2.0.0] ### feat - add detection for 20+ technology-specific rules including Tailwind CSS, shadcn/ui, Supabase, TypeScript - add automatic package manager detection (pnpm, yarn, npm, bun) based on lock files - add build tool recognition (Turbopack, Vite, Next.js) with version-specific features - add IDE detection for VS Code, Cursor, Windsurf, JetBrains, Zed - add library-specific guidelines for UI libraries like shadcn/ui and Radix UI - increase rule limit from 10 to 20 for better technology coverage - add real script extraction from package.json - add AI Assistant field to generated configurations ### fix - resolve GitHub Copilot adapter undefined property errors with optional chaining - improve framework and library matching with better aliases - fix library guidelines appearing in CLAUDE.md files - add graceful degradation for edge cases and invalid paths ### perf - improve relevance scoring algorithm with platform-specific filtering - improve content processing with mobile pattern exclusion - improve template processing for actionable guideline extraction - add error recovery for missing dependencies ### docs - add comprehensive MDX documentation system - add blueprint specifications for all rule formats - add IDE configuration reference guides - add GitHub repository documentation templates ### refactor - improve project documentation structure - add integration guides with practical examples ### [1.0.0] ### feat - add project analysis and pattern detection engine - add multi-format rule generation (Markdown, MDC, XML, JSON) - add AI assistant integrations (Claude Code, Cursor, Windsurf, GitHub Copilot) - add VDK Hub for team collaboration and rule sharing - add watch mode for continuous project monitoring - add comprehensive CLI with 20+ commands - add project templates for popular frameworks - add rule validation and error checking - add integration auto-detection and setup - add configuration management system ### feat - add project scanner with intelligent codebase analysis - add rule generator with context-aware rule creation - add integration manager for AI assistant setup and management - add hub client for team collaboration through rule sharing - add watch system for real-time monitoring and automatic updates - add validation engine for rule and configuration validation ### feat - add support for frontend frameworks (React, Vue, Angular, Svelte, Next.js, Nuxt.js, SvelteKit) - add support for backend frameworks (Node.js, Express, FastAPI, Django, Rails) - add support for languages (TypeScript, JavaScript, Python, Go, Rust) - add support for tools (Tailwind CSS, Prisma, tRPC, GraphQL, Docker) ### feat - add team creation and management - add rule sharing and discovery - add version control and rollback - add public and private deployments - add search and filtering - add analytics and usage insights - add collaborative rule development --- _Generated using [ai-changelog-generator](https://github.com/entro314-labs/AI-changelog-generator) - AI-powered changelog generation for Git repositories_