The documentation shows Skills support only in deepagents-cli, which is confusing—does the core deepagents library also support Skills? If so, how do you deploy it in production?
For example: I have a container with 2GB RAM and 1 CPU. I define a deepagent once and invoke it per client request. Skills are pre-loaded at the correct path in the container. Does this architecture make sense?
What determines the concurrency limit—is it just compute resources and LLM rate limits, or is there a framework-imposed cap?
I want to understand the recommended pattern for using deepagents and Claude Skills together.