# Log 2023/08/2023-08-08
- [[RWISE - 2023-08-07 - Edsger Dijkstra’s One-Day Workweek]]
- [[RWISE - 2023-08-06 - Interview With Audrey Watters About ‘Teaching Machines’]] - still need to finish this one, it's good
## Claude for comments on code
In [[2023-07-26]] I was mentioning using claude to generate more cool examples, so I decided to try it out, and by accident pressed enter too early. It ended up giving me some comments about the rename-column middleware in [[PROJ - GO GO GOLEMS - GLAZED]] that are actually really useful:
> [!ai]+ Claude
> This implements a nice middleware for renaming columns in structured data using both direct mappings and regular expressions. Here are a few thoughts:
> - The code overall looks well structured and easy to follow. Using a middleware pattern makes sense for this type of reusable data transformation.
> - Storing the renamed columns in a map is a good optimization to avoid repeatedly applying the same regexes.
> - For the regex renaming, it might be good to support ordered replacements rather than just taking the first match. This would allow chaining a series of regex replacements.
> - The YAML parsing for the regex config could be simplified a bit using a custom UnmarshalYAML method rather than generic map parsing.
> - For usability, it could be good to support column mappings in the CLI directly rather than needing a config file.
> - The documentation provides nice examples of how to use the renaming features. Adding a section on the config file format could be helpful.
> - One additional feature could be to support renaming columns based on their position rather than name, for cases where the header names are missing or untrusted.
> Overall the implementation looks solid, with a few potential enhancements to make it more usable. The middleware approach is clean and enables reuse across commands. Nice work!
This is really cool and something I should add to my workflow. Potentially I can couple this with a diff, use the diff to extract functions/files/methods that changed, and then use that in combination with oak to query for comments…
- [[PROJ - GO GO GOLEMS - OAK#diff + oak for pull request feedback]]
Prompting it correctly, it gave me one interesting idea, which is to ellipsize column names, which made me realize that I don't necessarily need the model to give me the proper command line from just the code (although that might work), but that for some help pages, it really matters to just ideate on example ideas, not the exact example. In fact, ideating on possible applications would help to design the feature up front, such as the suggestions above, which are really useful. This could also just be part of a process to create tickets themselves, and part of the RFC writing process.
- create a semi-correct prototype using code gen, doesn't need to be perfect, but at least embody the concept of the implementation
- have the prototype (generated or otherwise) be reviewed by the LLM before making it part of the actual RFC spec
- See [[IDEA - GO GO GOLEMS - Idea repository#Add git support to oak]]
## Processing [[SKETCHBOOK - 2023-05-29 - weinheim sketchbook]]
I am going through pages 8 and 9 which have the initial ideation for the context manager.
- [[IDEA - GO GO GOLEMS - 2023-08–05 - Prompt Context Manager]]
I am this time taking a picture of the page before the rework and one after, and writing down not just the ideas on the page (and trying to put them in context of the new work I've done), but of course also continuing to ideate. For that purpose, I'm adding [[IDEA - GO GO GOLEMS - Idea repository]].
- [[PROJ - GO GO GOLEMS - OAK#Tree like output structure]]
- [[IDEA - GO GO GOLEMS - Idea repository#Greasemonkey with LLM driven CSS/HTML query and modification]]
- [[IDEA - GO GO GOLEMS - Idea repository#HTML and CSS querying in oak]]
- [[IDEA - GO GO GOLEMS - Idea repository#Log querying tool]]
I also wrote up this process for mastodon:
Turning it into a blog post:
- [[DRAFT - 2023-08-08 - My software design process]]
And a little thread about working for small companies, which I guess I could turn into a structure note:
## Ideas for [[PROJ - GO GO GOLEMS - MASTOID]]
It would be great to have the format of the rendering from mastoid be usable for a blog article. I think maybe the best way to do that is to use the TemplateLookup mechanism from parka and provide a few default templates, and the ability for users to provide not just their own templates, but their own template directories, to make it easier to do split out templates. Maybe not entirely necessary for mastoid, but maybe a GO GO TEMPLATES project would be good to have, seeing how important template processing is across the board. This could also take care of the visual glamour output and other features, maybe by using a sub project to avoid pulling in glamour dependencies for WASM based projects (the thing that parses the terminal size, etc…) Maybe that's something I can actually disable with a config flag.
## RWISE tasks
(description includes #toRead)
created after last month
sort by created reverse
(description includes #toFile) OR (description includes #toProcess)
created after last month
sort by created reverse