Fully embracing separation of agent/informant "recipes" and target tbls

[jl5000]

I have a use case where we have a master dataset containing all columns and rows. This data is then used for bespoke downstream analyses, often using subsets of this data. I would like to use the same validation plan and data dictionary on these smaller datasets.

The functionality of pointblank already goes some way in allowing this re-use, through functions such as has_columns(), set_tbl(), and yaml_write(), however I don't think the package has yet fully embraced this approach.

I have made a couple of suggestions below which would make this use case fully possible, it just depends on whether it fits with your vision:

1. If set_tbl() allows you to change the target dataframe, then is there any need to define the tbl when creating the agent/informant? It would allow the full separation between "recipe" and tbl. As far as I can see interrogate()/incorporate() already provides informative errors when the tbl defined isn't valid.

2. When writing an agent to YAML (pre-interrogation), the package does a great job in preserving all of the logic in preconditions and active, such that it can be re-used downstream. However, this isn't the case for informants where info_columns() and info_snippet() don't seem to be recorded.

Recording the "recipes" for these objects to YAML is a really useful thing, and would negate the need to create a separate package containing functions to regenerate the agents/informants.