What problem does it solve? Getting a second opinion from a frontier model on a codebase usually means manually copying files into a chat window, losing context, and wasting tokens. The oracle CLI bundles your prompt plus selected files into a single one-shot request so another model can answer with real repository context, via API or browser automation. ## Core Features & Use Cases - Prompt + File Bundling: Attach files, directories, and globs with --file, including negation patterns, .gitignore awareness, and automatic exclusion of node_modules, dist, and files over 1 MB. - Browser and API Engines: Run long-thinking sessions with GPT-5.2 Pro in ChatGPT via --engine browser, or use --engine api for Claude, Grok, Codex, and multi-model runs. - Session Management: Long browser runs are stored under ~/.oracle/sessions; list them with oracle status and reattach with oracle session <id> --render instead of re-running. - Dry-Run Previews: Estimate payload size and token spend with --dry-run summary and --files-report before spending anything. - Use Case: You hit a stubborn bug in a TypeScript service. Run oracle --engine browser --model gpt-5.2-pro -p "<task>" --file "src/**" to get a deep, context-aware analysis from GPT-5.2 Pro, then verify the advisory output against your code and tests. ## Quick Start Ask the assistant to run oracle with the browser engine and GPT-5.2 Pro, attaching your src directory, to get a second opinion on your current bug or design question.