Feedback (SA team, re: @yongliang Xie): the eval skills require too many API keys; configuring them all up front is a real barrier. Proposal: a unified key + self-service provisioning.
目前 eval 这个类型的 skill，对 api key 要求比较高，一次性要配置那么多 key 本身就是很难完成的一个任务。从内部角度看，是否 @yongliang Xie 可以出一个统一的 key，后续想要去执行的用户就从你这里申请即可。当然更好的是自动化，比如用户可以在 agent 内提交自己的 bytedance 邮箱申请这个 key，然后 agent 中回复生成的 key 以及 scope 范围，接着就可以使用了
Why it hurts today: running the eval family can require several provider keys — e.g. BYTEPLUS_API_KEY, GOOGLE_AI_STUDIO_API_KEY / Vertex creds, OPENAI_API_KEY, AIDP_API_KEY, LLM_API_KEY, OPTIMIZER_MODEL_ID. Getting all of them configured before anything runs is a high bar.
Proposal:
- Short term: one internal unified key, requested from @yongliang Xie.
- Better: self-service in-agent — user submits their ByteDance email, agent returns a scoped key ready to use.
Feedback (SA team, re: @yongliang Xie): the eval skills require too many API keys; configuring them all up front is a real barrier. Proposal: a unified key + self-service provisioning.
Why it hurts today: running the eval family can require several provider keys — e.g.
BYTEPLUS_API_KEY,GOOGLE_AI_STUDIO_API_KEY/ Vertex creds,OPENAI_API_KEY,AIDP_API_KEY,LLM_API_KEY,OPTIMIZER_MODEL_ID. Getting all of them configured before anything runs is a high bar.Proposal: