Idea for a "Metatool" to generate Model Context Protocol servers, given specifications of existing tools, in order to be able to consume them in CursorAI, GitHub/Actions or in CLI prompting
#Q when should we encourage SMEs to add to a distinct LangSmith/Dataset vs when to apply a #Rubric ?
The annotation queue has a "D" hotkey for adding to the default dataset, so it might be simplest if we were going to only use datasets for the "golden" dataset - the "reference" of a good response.
Creating LangSmith/Annotation/Rubric
we can leave high level instructions, e.g. "Please only score fields you are comfortable with"
full list of feedback tags from our LangSmith/Workspace
Langsmith Online Evaluators provide a way in the langsmith UI to define a prompt and a Structured Output schema, and ensure that every time a trace comes in that matches a LangSmith/Filter, it gets classified
The main idea of mise is that mise/Config/mise.toml can let either a project or a user declare the
mise.toml describes consistent way to setup and interact with projects across languages
mise.toml is a file type used by the mise cli that lets developers globally declare or individual projects locally declare the programming languages, the package manager packages, the environment variables, and the common tasks that need to be run
AFAIK there's no dedicated global upgrade subcommand that just upgrades global mise tools, but there might be a way to call a command only on a particular (global) mise/Config/mise.toml
lets you specify configuration like AWS_ACCESS_KEY_ID that may differ between projects. It can also be used to automatically activate a Python virtualenv when entering projects too.