Open petruut opened 3 months ago
not for everyone, but this macOS app seems to embed fabric in a nicely integrated app https://alterhq.com/demos
What if we generate a table with pattern descriptions and feed that into a pattern to suggest which one to use? We could generate the descriptions via LLM using a summarize pattern pattern.
Pseudocode:
foreach pattern:
cat {pattern} | fabric summarize-pattern >> pattern-descriptions.csv
echo "You are a XYZ, suggest a pattern from the following table:" >> {patterns-dir}/select-pattern/system.md
cat pattern-descriptions.csv >> {patterns-dir}/select-pattern/sytem.md
Then fabric suggest-pattern -t "I want to summarize last quarter's report as a series of cat memes"
What do you need?
It's a great idea to have a library of patterns - however, in a long list without any qualification, it's difficult to find the right use cases.
A UI to use and identify the right patterns would solve this. We need to simplify the heuristics.