Zedong-Liu/codex-deepseek-sidecar — reverse-engineered prompt

Reverse engineered prompt

GitHub

Please set this repo up so my coding assistant can use DeepSeek sidecar workers while the main model stays in charge. I want the stronger model to handle planning, judgment, and the final answer, and send long tests, CI log digging, broad codebase exploration, isolated reviews, and implementation attempts to cheaper DeepSeek workers inside the same normal workflow.

I already have a DeepSeek API key. If it is not configured on this machine yet, ask me for it. If needed, set up the local proxy and any profile or adapter settings, start the built in proxy, and keep it running so the sidecars work reliably. Then launch one DeepSeek sidecar for this repo and make sure later requests can return to the right worker session.

When you are done, use it on one real task, like a slow test run or a log analysis pass, and give me the combined result from the sidecar and the main model. Check the repo docs and current docs online if anything is unclear.

Want more depth? Deep Reverse