it's really magic
it's really magic
i want to define multiple versions of a skill, acceptance criteria (to be evaluated by a separate agent), and a runner that does repeated runs of different versions of my skill to see if there’s a statsig improvement in any version
i want to define multiple versions of a skill, acceptance criteria (to be evaluated by a separate agent), and a runner that does repeated runs of different versions of my skill to see if there’s a statsig improvement in any version
well i think part of the reason LLMs don't do a great job in complex brownfield projects is because they don't understand past design choices and constraints that aren't directly expressed in code
so i'm wondering if making those queryable can help. it helps humans!
well i think part of the reason LLMs don't do a great job in complex brownfield projects is because they don't understand past design choices and constraints that aren't directly expressed in code
so i'm wondering if making those queryable can help. it helps humans!
i've made a claude skill that uses @bobbby.online's deciduous to generate a design evolution tree from a project's commit history. this tree is generated by Claude using the displayed prompt (and my skill). the graph ~matches how i remember it
react-deciduous-example.pages.dev
i've made a claude skill that uses @bobbby.online's deciduous to generate a design evolution tree from a project's commit history. this tree is generated by Claude using the displayed prompt (and my skill). the graph ~matches how i remember it
react-deciduous-example.pages.dev
npx woodshed
npx woodshed
I just compiled the Less.js source code into a C library. Then I called it from Rust as a native Parcel plugin (no Node). Wild. 👨🔬
devongovett.me/blog/static-...
I just compiled the Less.js source code into a C library. Then I called it from Rust as a native Parcel plugin (no Node). Wild. 👨🔬
devongovett.me/blog/static-...
even if there’s a winning wiring of smarts and duct tape, clearly it’ll take a few months for someone to package it up in a product. then competition will sort it out, then open source will copy it half-badly, then it will git good.
why bother w/ middle steps?
even if there’s a winning wiring of smarts and duct tape, clearly it’ll take a few months for someone to package it up in a product. then competition will sort it out, then open source will copy it half-badly, then it will git good.
why bother w/ middle steps?
www.noaa.gov/news-release...
it lets you interactively step through how the React Server Components protocol works directly in the browser — no server needed.
it lets you interactively step through how the React Server Components protocol works directly in the browser — no server needed.
would anyone find it useful?
would anyone find it useful?
lots of bugfixes and a few new examples.
this basically works now! i'll open source a bit later
lots of bugfixes and a few new examples.
this basically works now! i'll open source a bit later
the stack:
- atproto (jetstream + new lex client)
- next 16 + cacheComponents (yes, a recent version)
- postgres + drizzle, redis
- vitest
the stack:
- atproto (jetstream + new lex client)
- next 16 + cacheComponents (yes, a recent version)
- postgres + drizzle, redis
- vitest
did -> handle, handle -> did, and did -> avatar lookups should all be getting cached in redis now which should considerably speed up navigations and stop 1s+ load times on some pages
cause was fun: github.com/vercel/next....
did -> handle, handle -> did, and did -> avatar lookups should all be getting cached in redis now which should considerably speed up navigations and stop 1s+ load times on some pages
cause was fun: github.com/vercel/next....
wtbb.vercel.app
I tried browsing the React repo example and it's _fast_! Really impressive!
wtbb.vercel.app
I tried browsing the React repo example and it's _fast_! Really impressive!