Raycast-PromptLab

Resolves template placeholders ({{selectedFiles}}, {{clipboardText}}, {{todayEvents}}, {{currentApplication}}) at runtime by querying macOS system APIs, Raycast context, and file system state. Uses a placeholder resolution pipeline that maps placeholder tokens to resolver functions that fetch real-time context data, enabling prompts to dynamically bind to user environment state without manual context passing.

Executes AppleScript or shell commands after AI response generation, enabling post-processing automation workflows. Parses action script definitions from command configuration, executes them in the system shell or AppleScript runtime, and chains results back into the conversation context. Supports conditional execution based on AI response content and error handling with fallback behaviors.

Extracts context from the active browser tab including page title, URL, selected text, and full page content. Injects browser context into prompts via placeholders like {{browserTabTitle}}, {{browserTabURL}}, and {{selectedBrowserText}}. Enables AI commands to analyze web content, summarize articles, and answer questions about the current webpage without manual copy-paste.

Provides granular configuration options for command behavior including temperature, max tokens, system prompts, timeout settings, and response formatting. Stores settings in Raycast preferences, enabling users to fine-tune AI model behavior and command execution without modifying command definitions. Supports per-command overrides of global settings.

Supports importing and exporting command definitions as JSON files, enabling backup, migration, and sharing of command configurations. Implements JSON serialization of command metadata, prompts, action scripts, and settings. Provides import validation to detect incompatible command versions and handles data migration when PromptLab updates change the command schema.

Implements a searchable command palette (search-commands.tsx) that allows users to quickly find and execute PromptLab commands by name, description, or tags. Provides fuzzy search matching, command preview, and one-click execution. Integrates with Raycast's command search to make PromptLab commands discoverable alongside native Raycast commands.

Provides a menubar item that offers quick access to frequently-used PromptLab commands without opening Raycast's main window. Allows users to pin commands to the menubar for one-click execution. Displays command status and recent results in the menubar dropdown, enabling rapid command invocation from anywhere on macOS.

Abstracts AI model interactions behind a unified interface supporting OpenAI, Anthropic, and custom HTTP endpoints. Manages model configuration including API keys, base URLs, and request/response schemas. Implements request marshaling that converts PromptLab command context into model-specific input formats and parses model-specific response structures back into unified conversation objects.

Implements a multi-turn chat interface (CommandChatView) that maintains conversation history and enables autonomous agent behavior where the AI can invoke other PromptLab commands to fulfill user requests. Manages conversation state, renders streaming responses, and provides a command invocation mechanism that allows the AI to recursively call other commands with context from the current conversation.

Extracts and injects content from selected files into prompts, supporting multiple file types including text, code, images, audio, video, and office documents. Implements content extraction pipelines that parse file formats (PDF text extraction, image OCR, audio transcription, video frame analysis) and inject extracted content as context into AI prompts. Handles binary file formats and large files with size-aware truncation.

Provides a visual command builder interface (create-command.tsx) that allows users to define custom AI commands without coding. Captures command metadata (name, description, icon), prompt template with placeholder syntax, model selection, action script definitions, and execution options. Stores command definitions as JSON in Raycast's local storage, enabling non-technical users to create reusable AI commands through a form-based UI.

Implements a command discovery interface (discover-commands.tsx) that connects to a community command store, allowing users to browse, preview, and import pre-built commands from other users. Manages command metadata including author, description, tags, and ratings. Supports command import/export via JSON serialization, enabling a marketplace-like ecosystem where users can share and discover reusable automation patterns.

Manages persistent variables that can be referenced across multiple command executions and conversations. Stores variable state in Raycast's local storage, enabling commands to maintain state between runs and share data across the command ecosystem. Supports variable interpolation in prompts using placeholder syntax and provides APIs for commands to read/write variables during execution.

Tracks execution history of all PromptLab commands, storing metadata including timestamp, command name, input parameters, AI response, and execution status. Provides a searchable history interface that allows users to review past executions, re-run commands with previous parameters, and export execution logs. Enables audit trails for compliance and debugging of automation workflows.

Integrates macOS speech recognition for voice input to commands and text-to-speech for reading AI responses aloud. Uses native macOS speech APIs to capture voice input, transcribe to text, and synthesize responses. Enables hands-free command execution and audio feedback, making PromptLab accessible for voice-driven workflows and accessibility use cases.