The lesson I took from building my own Skills library: the 80/20 isn't about which skills are best, it's about which ones you actually remember to invoke. A brilliant skill that lives in a menu you never open is worse than a mediocre one that auto-triggers on the right keyword. Design for invocation, not capability.
Do you use Anki? Flashcards give you a superpower to recall anything. So when you create a skill, just create (or have Claude Code create) a flashcard to remember what it does. That removes the ceiling on the # of useful skills you can have!
so i have to ai to ai to enable to ai to destroy me (us) because it’s ai 😂😂🫣🫣🫣 so far i remain truly unimpressed 😑😑😑
honestly (maybe not so much with the ai [it’s inert after all]) more with the people extolling its virtues/ all the while with the tacit threat that it’s going to be a technology that destroys us all, all obviously except those who extol it hey no fair 😑😑😑 😂😂😂 😑😑😑
(if You haven’t already)
just #read or #audiobook —
(not all #books just one #book [on the subject {if You know of others please ‘tag’ or ‘dm’ me with Your recommendation}])
The Decision Council skill doesn't work well because Claude isn't good at pretending to be someone else in the same thread. Works way better to delegate that to a subagent.
Try the Expert Review skill described here instead:
The lesson I took from building my own Skills library: the 80/20 isn't about which skills are best, it's about which ones you actually remember to invoke. A brilliant skill that lives in a menu you never open is worse than a mediocre one that auto-triggers on the right keyword. Design for invocation, not capability.
Do you use Anki? Flashcards give you a superpower to recall anything. So when you create a skill, just create (or have Claude Code create) a flashcard to remember what it does. That removes the ceiling on the # of useful skills you can have!
Testing 100 Claude skills to find the best ones is incredibly valuable. Saved this.
Such a valuable post! Thanks for sharing!!
I would also add that how you incorporate and build upon those skills can have a profound effect to improve them over time.
Just wrote about it yesterday at :
so i have to ai to ai to enable to ai to destroy me (us) because it’s ai 😂😂🫣🫣🫣 so far i remain truly unimpressed 😑😑😑
honestly (maybe not so much with the ai [it’s inert after all]) more with the people extolling its virtues/ all the while with the tacit threat that it’s going to be a technology that destroys us all, all obviously except those who extol it hey no fair 😑😑😑 😂😂😂 😑😑😑
(if You haven’t already)
just #read or #audiobook —
(not all #books just one #book [on the subject {if You know of others please ‘tag’ or ‘dm’ me with Your recommendation}])
TITLE
Al Snake Oil Al SnakeOil
AUTHORS
Arvind Narayanan & Sayash Kapoor
#ArvindNarayanan
#SayashKapoor
PUBLISHER
Princeton University Press
press.princeton.edu
@princetonupress
YEAR PUBLISHED
2024-09-24
visit:
https://www.normaltech.ai/
search:
YouTube - Arvind on AI
@ArvindOnAI
https://youtube.com/@arvindonaip
Skills feel like the missing layer between prompting and real workflows. This is where AI actually becomes useful.
The Decision Council skill doesn't work well because Claude isn't good at pretending to be someone else in the same thread. Works way better to delegate that to a subagent.
Try the Expert Review skill described here instead:
https://substack.com/home/post/p-193543324
Curious about how well the Figma skill works. Thanks for sharing:)