<h1 align="center">
<a href="https://prompts.chat">
All notable changes to this project will be documented in this file.
Sign in to like and favorite skills
All notable changes to this project will be documented in this file.
Linear Integration: Added comprehensive Linear support for accessing issues directly from the command line with rich formatting and full context.
vibe-tools linear connect - Interactive authentication setup supporting both personal API keys and OAuth2 PKCE flowvibe-tools linear get-issue <identifier> - Display detailed Linear issue informationvibe-tools linear issue <identifier> - Alternative command name for issue retrievalLINEAR_API_KEY in .vibe-tools.env filesITE-123) and UUID formatGitHub Command Improvements: Added GitHub command parameters (
--review-only, --discussion-only, --metadata-only, --no-links, --hide-resolved).
moonshotai/kimi-k2-instruct and qwen/qwen3-32b. Configure using the GROQ_API_KEY environment variable.
doppler CLI is on the PATH, loads any secrets that end with _API_KEY into the environment before commands execute. This means you no longer need to copy provider API keys into .env files for projects that already manage secrets with Doppler. The feature is enabled by default and can be disabled by adding "disableDoppler": true to your vibe-tools.config.json.--with-doc: You can now pass local files directly to the --with-doc flag (relative/absolute paths or file:// URLs) across the ask, repo, plan, and doc commands. This makes it simple to enrich prompts with Markdown, text, or other documents from your filesystem without hosting them online.--web flag for live web search with source citations and the --reasoning-effort flag (low, medium, high) for deeper analysis. Works across ask, web, repo, doc, and plan commands, bringing Grok feature-parity with Gemini and Perplexity providers.vibe-tools install command with intelligent configuration and rules file management.
Increased Default Max Tokens: Raised the default maximum output tokens from 8,000 to 24,000 across all commands for more comprehensive responses.
--max-tokens flag for specific needsUpdated Gemini Model Names to Stable Versions: Migrated all Gemini 2.5 model references from preview versions to stable versions throughout the codebase.
gemini-2.5-flash-preview-* → gemini-2.5-flashgemini-2.5-pro-preview* → gemini-2.5-proUpdated Default Models to latest versions:
gemini-2.5-flash-lite-preview-06-17 as a lightweight option.
CI=true, DEBIAN_FRONTEND=noninteractive, or NONINTERACTIVE=true environment variablesCURSOR_TRACE_ID environment variable is presentvibe-tools install) and JSON installation flowsSKIP_PLAYWRIGHT=1 environment variablenpx playwright install commandsensurePlaywrightBrowsers() verification to all browser subcommands (open, act, extract, observe)--with-doc implementations with proper Playwright checks before browser automationvibe-tools browser open --connect-to with existing Chrome instances.
--connect-toBrowser Mac-Chrome Lite Mode: Added
--lite flag to vibe-tools browser mac-chrome command for faster startup and minimal side-effects.
Git Diff Integration for Repo Command: Added
--with-diff flag to vibe-tools repo command for enhanced code review capabilities.
--base flag to specify different branch for diff comparison--with-doc flag for comprehensive analysisvibe-tools browser mac-chrome command to start a Chrome instance with remote debugging on macOS.
claude-3-7-sonnet-latest, o3, o4-mini, gpt-4.1, gpt-4.1-mini, gemini-2.5-flash-preview-04-17, and groq-llama-3.3-70b-versatilemodelClientOptions for better compatibility--web flag to enable web search capabilities for supported models (currently Gemini models) across all commands.claude-3-7-sonnet-latest to claude-sonnet-4-20250514--debug flag and during token limit errors.--with-doc URLs for ask, repo, plan, and doc commands, allowing users to provide context from several documents simultaneously.Doc & Browser Commands Provider Improvements:
CTA improvement:
--with-doc <doc_url> flag to repo and doc commands to fetch content from any web URL and include it as context.
--with-doc <doc_url> flag to repo and doc commands to fetch content from any web URL and include it as context.
codex.md at repository root and global ~/.codex/instructions.mdgrok-3-latest, grok-3-mini-latest). Configure using the XAI_API_KEY environment variable.--with-doc <doc_url> flag to repo and doc commands to fetch content from any web URL and include it as context.Youtube command: Added
youtube command that can be used to analyze YouTube videos and generate detailed reports:
GEMINI_API_KEY to be set in your environment or .vibe-tools.env fileSupport for Reasoning Effort Parameter: Added
--reasoning-effort parameter to enhance the quality of responses for complex queries:
low, medium, and high values for controlling reasoning depthOVERRIDE_SAFETY_CHECKS=true environment variablesrc/commands/test/command.ts where test scenario details were not being correctly displayed in generated reports. The fix ensures that the promises returned by queue.add() are properly handled, allowing detailed scenario information to appear in test reports.--from-github parameter to the repo command, enabling analysis of remote GitHub repositories without local cloning. This feature provides the same functionality previously available in the doc command, making it easier to get context-aware assistance for any public GitHub repository (e.g., vibe-tools repo "explain the authentication flow" --from-github=username/repo-name).Support for Custom Repomix Configuration: Added support for
repomix.config.json files to customize which files are included/excluded during repository analysis. This configuration file can be placed in the repository root and will be automatically detected by repo, plan, and doc commands, allowing for more precise control over repository content analysis.
Subdirectory Analysis for Repo Command: Added
--subdir parameter to the repo command, allowing users to analyze a specific subdirectory instead of the entire repository. This makes the repo command more efficient when working with large codebases by focusing only on relevant subdirectories (e.g., vibe-tools repo "explain the code" --subdir=src/commands).
Improved Model Name Resolution: Enhanced model name handling across providers to better handle experimental and latest model versions:
-exp-* suffixes to find stable model versions-latest suffixes to the most recent compatible modelgpt-4o will find openai/gpt-4o)Support for OpenRouter when using MCP
By default the
mcp command uses Anthropic, but now takes a --provider argument that can be set to 'anthropic' or 'openrouter'
MCP commands require
ANTHROPIC_API_KEY or OPENROUTER_API_KEY to be set in your environment
The '--model' arg can also be used with the MCP command to set the OpenRouter model. It is ignored if the provider is Anthropic.
Google Vertex AI Authentication: Added support for Google Vertex AI authentication using JSON key files or Application Default Credentials (ADC). This update maintains backward compatibility, continuing to support the direct API key string method, while adding additional authentication options. To use this feature, set the
GEMINI_API_KEY environment variable to the path of your JSON key file or to adc to use Application Default Credentials. This enables access to gemini models such as gemini-2.0-flash via the Vertex AI. This feature introduces a new dependency: google-auth-library.
GEMINI_API_KEY to the path of your service account JSON key file:
GEMINI_API_KEY="./path/to/service-account.json"
Then setgcloud auth application-default login
GEMINI_API_KEY to adc to use Application Default Credentials:
GEMINI_API_KEY="adc"
New Test Command: Added comprehensive test command for automated feature behavior testing:
Improved ModelBox Provider: Enhanced the ModelBox provider with improved model name handling. If a requested model is not found, vibe-tools now provides helpful suggestions for similar models. Error messages have also been clarified to better guide users on the requirement for provider prefixes when specifying ModelBox models.
Example: Improved Error Message If you use an invalid model with ModelBox, you will now receive suggestions:
Error: Model 'invalid-model' not found in ModelBox. You requested: invalid-model Similar available models: - openai/gpt-4o - anthropic/claude-3-5-sonnet Use --model with one of the above models. Note: ModelBox requires provider prefixes (e.g., 'openai/gpt-4' instead of just 'gpt-4').
Added support for MCP server overrides in the marketplace
MCP_OVERRIDES mapgithubUrl, command, and argsvibe-tools.config.json
mcp.overrides sectionclaude-3-7-sonnet for most use cases and claude-3-7-sonnet-thinking for MCP clientp-queue for parallel execution managementglob and @types/glob for test file pattern matchingplaywright and playwright-core with version pinning--connect-to:
reload-current as a special URL value to refresh the current page without losing the connected sessionopen, act, observe, extract) now have --console and --network options enabled by default. Use --no-console and --no-network to disable them.--connect-to: now reuses existing tabs instead of creating new ones for better state preservation.cursorrules or new .cursor/rules/vibe-tools.mdc)USE_LEGACY_CURSORRULES environment variable to control cursor rules file locationask command to pass debug flag to providerweb command to properly handle debug flagdoc command to include debug flag in optionsplan command to pass debug flag to both file and thinking providersask command for direct model queries
.cursor/rules/vibe-tools.mdc by default.cursorrules file via USE_LEGACY_CURSORRULES=trueUSE_LEGACY_CURSORRULES environment variablegpt-4o model in browser commands (act, extract, observe)
--model=gpt-4o command-line optionvibe-tools.config.jsonplan command with dual-provider architecture:
fileProvider handles repository file analysisthinkingProvider generates implementation plans--fileProvider: Provider for file identification (gemini, openai, or openrouter)--thinkingProvider: Provider for plan generation (gemini, openai, or openrouter)--fileModel: Model to use for file identification--thinkingModel: Model to use for plan generation--fileMaxTokens: Maximum tokens for file identification--thinkingMaxTokens: Maximum tokens for plan generation--quiet flag to suppress stdout output while still saving to file with --save-to
--connect-to by not forcing a specific color scheme--connect-to--connect-to without an explicit --viewport optionopen, act, observe, extract) now have --console and --network options enabled by default. Use --no-console and --no-network to disable them.--connect-to instead of creating new ones--connect-to unless --viewport is explicitly provided--video with --connect-to as video recording is not supported when connecting to existing Chrome instancesreload-current as a special URL value for browser commands when using --connect-to. This allows refreshing the current page while maintaining the connection, which is particularly useful in development workflows.--url=current to skip navigation and use the current pagebrowser command for AI-powered web automation and debugging, leveraging Stagehand AI for natural language interaction
act <instruction> --url <url> [options]: Execute actions on a webpage using natural language instructions
<instruction>: Natural language instruction describing the action (e.g., "Click Login", "Type 'hello' in the search box")--url <url>: Required URL to navigate to before performing actions--delay=<milliseconds>: Delay between actions (default: 100)--retry=<number>: Number of retries for failed actions (default: 3)extract <instruction> --url <url> [options]: Extract data from webpages based on natural language instructions
<instruction>: Natural language instruction describing the data to extract (e.g., "product names", "article content")--url <url>: Required URL to navigate to--format=<json|csv|text>: Output format (default: json)--flatten: Flatten nested objects in outputobserve <instruction> --url <url> [options]: Observe interactive elements on a webpage and suggest possible actions
<instruction>: Natural language instruction describing what to observe (e.g., "interactive elements", "login form")--url <url>: Required URL to navigate to--interval=<milliseconds>: Check interval for observation (default: 1000)--duration=<duration>: Total observation time (e.g., '30s', '5m')--changes-only: Only report when changes are detectedopen <url> [options]: Open and interact with web pages
--html--console--network--screenshot=<file path>--viewport=<width>x<height>--no-headless--connect-to=<port>--timeout=<milliseconds>: Set navigation timeout (default: 30000)--viewport=<width>x<height>: Set viewport size (e.g., 1280x720)--headless: Run browser in headless mode (default: true)--no-headless: Show browser UI for visual inspection and debugging--connect-to=<port>: Connect to an existing Chrome instance--wait=<duration or selector>: Wait after page load, supports:
punycode and node:punycode imports to punycode/o200k_base for better compatibility with Gemini modelstokenCount.encoding in config filedoc command
@branch syntax (e.g. --fromGithub=username/repo@branch)ref parameterdoc command to generate comprehensive repository documentation
--fromGithub option--outputpnpm dev for running latest code without building.cursorrules to include documentation for the new doc commandInitial release with basic functionality:
--with-doc <doc_url> flag to repo and doc commands to fetch content from any web URL and include it as context.youtube command for analyzing YouTube videos (summaries, transcripts, etc.).