oracle

Oracle is a specialized CLI utility designed to bridge the gap between local software development environments and large language models. It enables developers to curate specific file sets, directories, or globs to create high-context 'one-shot' requests. By automating the extraction and preparation of project context, Oracle ensures that AI models receive the necessary information—such as stack architecture, build tools, configuration files, and specific error logs—required to generate accurate, actionable solutions for debugging, refactoring, or architectural design reviews. The tool is particularly optimized for long-thinking models like GPT-5.2 Pro, facilitating deep, multi-step analysis within a managed session lifecycle.

• Streamlines prompt engineering by automatically aggregating relevant codebase context based on user-defined file patterns and glob filters.

• Supports dual-mode execution: browser-based automation for long-context tasks with persistent session support, and standard API mode for programmatic or multi-model workflows.

• Includes robust preview and reporting tools, allowing users to verify payload size and token consumption before sending requests, preventing unnecessary costs.

• Features a session management system that allows users to reattach to long-running analysis tasks if a connection drops or a timeout occurs.

• Respects standard development environment conventions, including .gitignore, while allowing fine-grained control over included/excluded directories and file types.

• Always include a comprehensive project briefing within your prompt, detailing the stack, entry points, and desired output format (e.g., patch plans, unit tests).

• Use the --dry-run and --files-report flags to audit token usage and confirm the correct file set before executing the main query.

• The tool ignores common build artifacts like node_modules and coverage folders by default, ensuring the AI focuses on source code logic.

• For complex investigations, utilize the 'Exhaustive Prompt' pattern by providing clear repro steps, exact error text, and clear constraint boundaries.

• Oracle is a one-shot context tool; ensure that all necessary entry points and configuration files are explicitly defined in the file set for the best model performance.