Hey ProductHunt community! 👋 As a developer working with LLMs daily, I noticed a huge issue: dumping an entire repository into Claude or ChatGPT Canvas ruins the model's accuracy due to the 'Lost in the Middle' effect, and it completely drains your token budget. Standard tools act like dumpsters — they just pack everything. I wanted a scalpel. That’s why I built ContextSniper. It runs 100% locally, parses your import chains to build a dependency graph, and uses BM25 scoring to extract only the files needed for your specific task (e.g., npx context-sniper "fix auth middleware"). If the code exceeds your budget, it automatically strips files down to class/method signatures so you never overflow your limits. It's fully open-source and ready to run via npm. Would love to hear your feedback!

ContextSniper

Smart task-driven LLM context optimizer

Smart task-driven LLM context optimizer