What problem does it solve?
It solves the problem of getting useful, repository-grounded answers by bundling a targeted set of files with your prompt into a one-shot request that preserves real context.
Core Features & Use Cases
- Best-practice one-shot bundling: Combines your prompt with carefully selected repo files for a model to answer with concrete context.
- Engine workflows (browser vs API): Supports a long-running browser workflow for models like GPT-5.2 Pro and an API mode for other provider needs.
- Token- and session-aware execution: Uses dry-run previews, file reports, and reattachable sessions to avoid wasted reruns and to manage timeouts.
- Safe attachment patterns: Applies include/exclude globs, respects .gitignore, filters dotfiles by default, and rejects overly large files.
Quick Start
Run oracle with browser engine mode, attach the smallest file set that contains the truth, and use the generated one-shot context for an advisory repo-grounded answer.