With most software projects, the repo is the source of truth. You can clone it, inspect the code/configs, and understand how the system works. But with agents, a lot of the actual behavior lives outside the repo. Prompts, framework-specific configs, tool wiring, memory setups, hosted dashboards, and other glue code end up scattered everywhere. And portability still feels messy. Move to a different framework or model and things start behaving differently in ways that are hard to track or version properly.

Feels like the ecosystem still hasn’t figured out a clean git-friendly way to structure agent projects so they’re actually reproducible from the repo itself.

Curious how people here are dealing with this right now.

[留言]

为什么值得关注

能改变理解方式，而不只是重复常识；符合当前抓取需求；它提供了新的理解或解释，而不只是表面观点

来源：reddit，领域：tech，保留分：0.67