用 GPT-6 Astra 搭 AI 工程师的 8 个开源仓库
把 Serena、Context7、OpenHands、E2B 等 8 个开源项目串成流水线，让一个 GitHub issue 从理解到提 PR 全程自动跑完。
GPT-6 Astra + GitHub can replace a ridiculous amount of coding work 8 open-source repos for building an AI engineer that can enter a codebase and actually finish the job 01 Serena ▸ https://github.com/oraios/serena → map and navigate the codebase 02 Context7 ▸ https://github.com/upstash/context7 → pull current documentation 03 Repomix ▸ https://github.com/yamadashy/repomix → package the repo into useful context 04 LangGraph ▸ https://github.com/langchain-ai/langgraph → control the execution loop 05 OpenHands ▸ https://github.com/All-Hands-AI/OpenHands → turn tasks into actual code changes 06 E2B ▸ https://github.com/e2b-dev/E2B → run everything inside isolated sandboxes 07 SWE-bench ▸ https://github.com/SWE-bench/SWE-bench → test against real software issues 08 Langfuse ▸ https://github.com/langfuse/langfuse → trace what the agent did and why the build: issue → understand → retrieve → plan → code → execute → test → inspect → retry → PR the valuable part isn't having 8 AI tools open it's connecting them so one GitHub issue can travel through the entire stack without a human rebuilding the context at every step ⭣
这条是引用着另一条写的：x.com/promixyz/status/2102424036416348494