very useful skill by Shreya on plain writing.
I use similar skills inside Amazon for writing docs and narratives.
simple and clear writing is a service to readers;
only one person asked, but here is my "plain writing" skill: https://github.com/shreyashankar/plain-writing-skill
courtesy of our (ongoing) DocWriter project, which is teaching us about ai slopwords & how users fight them
Users thank the researcher for the Plain Writing GitHub Skill because it offers practical techniques to combat AI slop and improve writing clarity.
No Digg Deeper questions have been answered for this story yet.
@sh_reya i should have mentioned this before, but i've used your "Writing in the Age of LLMs" blog to give agents examples of bad writing and anti-patterns. so useful!
only one person asked, but here is my "plain writing" skill: https://github.com/shreyashankar/plain-writing-skill
courtesy of our (ongoing) DocWriter project, which is teaching us about ai slopwords & how users fight them
ai writing slop has moved on from em dashes to inventing hyphenated adjectives that no one has ever said before
@sh_reya Why not build the diff with a script instead of asking the LLM to create it
@sh_reya how does this compare to the humanizer skill?
https://github.com/blader/humanizer
@slimshetty_ Makes me so happy to hear!!
@sh_reya I didn't ask for it but I enjoyed a lot. Thank you for sharing
@sh_reya This is so good!
@sh_reya Lol i need this! those are literally same rules im using all the time!
@sh_reya would love your thoughts on how qvr could help maintain and audit skills as you develop DocWriter
https://github.com/astra-sh/qvr
@sh_reya Thanks Shreya
@sh_reya "Slopwords" is a great name for it — the words that show up when nothing was actually decided. Plain writing is mostly the discipline of pulling out exactly the right amount and stopping, which is hard precisely because the model never wants to stop.
@sh_reya Have you tried the humanizer skill? Btw I’m yoinking ur skill
https://github.com/blader/humanizer
Ha so far my subscriptions have been able to pay for the diff tokens…but yes a much more cost efficient design would be to use hooks to capture the writing and programmatically create the diff
Though for some formats, eg html, raw token diffs may not be great to read
Feel free to make a PR if you have a good idea 🤩
@sh_reya Wow.. Thanks.. I have been doing some work arounds myself like programatically identifying and feeding known slop patterns.. Your skill is genuinely useful. In your experience with docwriter do u get better results processing: pre slop gen or post slop gen?
@sh_reya the substitution arc keeps moving. em-dash to 'fundamentally' to hyphenated-mystery-adjectives. constraint prompts beat removal lists because removal pushes the model into nearby unclaimed territory.
