Building a coding loop for GPT-6 Astra with 7 open source repos
Aider, Serena, Context7, Playwright, E2B, pytest and Langfuse are chained into one pipeline, leaving Astra to decide what to change and write it.
GPT 6 Astra + GitHub is getting f…cking unfair 7 open source repos can turn one model into a full coding loop 01 Aider ▸ https://github.com/Aider-AI/aider → gives Astra the editing loop inside a real Git repo 02 Serena ▸ https://github.com/oraios/serena → finds symbols, references and the files connected to the task 03 Context7 ▸ https://github.com/upstash/context7 → pulls current documentation before the model touches an API 04 Playwright ▸ https://github.com/microsoft/playwright → lets the agent open the product and test what it actually changed 05 E2B ▸ https://github.com/e2b-dev/E2B → runs commands and code inside an isolated environment 06 pytest ▸ https://github.com/pytest-dev/pytest → checks whether the patch survives the existing test suite 07 Langfuse ▸ https://github.com/langfuse/langfuse → keeps the full trace when the run fails the build: → Serena finds the relevant code → Context7 pulls the right docs → GPT 6 Astra decides the change → Aider edits the repo → E2B runs it → pytest checks the code → Playwright checks the product → Langfuse records the whole run → PR I like this setup because the model never has to guess whether it finished the job And Astra stays focused on the only part worth giving to the model Figuring out what should change and writing it
This post quotes another post: x.com/promixyz/status/2102424036416348494