Large language models struggle with real codebases, hitting context limits with noisy, hierarchical files. A developer working on modern JS and Bun projects built srcpack, a CLI that generates LLM-optimized snapshots. It bundles code by domain—like web or API—with precise file and line number references, respecting `.gitignore` by default.

The tool outputs plain text files, making it easy to attach to ChatGPT, Gemini, or Claude. This approach avoids manual file pasting and helps explore large repositories, asking questions about auth or billing. It also serves as a lightweight, read-only AI interface to the codebase for non-technical teammates like PMs and designers.

While not a final solution for LLM-codebase interaction, srcpack offers a practical workflow improvement for daily development. It addresses a common pain point as projects scale, where context becomes the bottleneck. The tool is available via `npx srcpack` or `bunx srcpack` with zero configuration needed.