Kasper Peulen

🤟TypeScript fanatic | ❤️ building apps with React + Storybook | Storybook maintainer @chromaui

[Kasper Peulen @KasperPeulen]

@mattpocockuk Mid task compacts in a goal are not a problem as long as you have a clear spec issue listed in the goal text and clear acceptance criteria

[Kasper Peulen @KasperPeulen]

@mattpocockuk @LukeberryPi grill-with-docs is too tightly coupled
We have our docs website hosted on `docs`.
It should just find the current conventions for where things live
/setup-matt-pocock-skills is a skill smell!
Really, it can not deduce itself that I use github??!?

[Kasper Peulen @KasperPeulen]

@mattpocockuk Do you have any proof that for new models such as Sol and Fable the dumb zone start at 150k tokens.
My main experience is that, the more context I give them, the better the results are. Things like discussion transcripts, slack threads etc.

[Kasper Peulen @KasperPeulen]

@mattpocockuk I don't agree. I think you underestimate how "much context/effort" a good discussion or implementation with an agent often requires in big long lived codebases/companies.

[Kasper Peulen @KasperPeulen]

@andrestaltz The problem is though.
You can spend $6000 dollars worth of API tokens in Codex or Claude for the price of $200 monthly subscription.

[Kasper Peulen @KasperPeulen]

@mattpocockuk I use something similar, but it is more like:
Grill -> High level issues (6 week plan)

Then for each issue, I grill it again before implementing, because things change all the time:
Grill -> Update issue -> Goal -> Review -> Merge

[Kasper Peulen @KasperPeulen]

@emanueledpt @cheepo2109 @trySynara But it does not have connectors (Slack, Linear etc)?

[Kasper Peulen @KasperPeulen]

@mattpocockuk I wonder if there is any evidence that this dumb zone is really this bad still with new models. I have had grilling sessions taking taking many hours. I let the agent record all decisions, and research sources in a md file, so it would survive compacting. This seems to work fine.

[Kasper Peulen @KasperPeulen]

@mattpocockuk @kentcdodds Some might say that is a premature optimization 😅 I think there is something to say for using docs/.md files over skills for many cases!

[Kasper Peulen @KasperPeulen]

@mattpocockuk @kentcdodds What is the difference with using @ to find them though?

[Kasper Peulen @KasperPeulen]

@steipete @CongGangster Every model is SO bad in writing skills. Or AGENTS.md files. GPT 5.5 may be the worst.

[Kasper Peulen @KasperPeulen]

@thsottiaux /goal

[Kasper Peulen @KasperPeulen]

@Dimillian Git diff for last turn, uncomitted, branch etc! And being able to write comments!

[Kasper Peulen @KasperPeulen]

Check it out!
https://t.co/xKo9b3H2oj

[Kasper Peulen @KasperPeulen]

vitest-plugin-rsc 0.2.0 is released and now runs all public Next.js APIs through real Next.js code paths!

[Kasper Peulen @KasperPeulen]

Excited about this: got a lightweight async_hooks compatibility shim working in the browser for vitest-plugin-rsc.

This should allow us run Next’s real cookies() and headers() request-context code in browser-mode RSC tests, without mocking those APIs.

@kettanaito

[Kasper Peulen @KasperPeulen]

@kettanaito Nice, is this also available in vitest browser mode?

[Kasper Peulen @KasperPeulen]

@kettanaito Yes, let's explore a node based and browser based RSC approach in parallel for vitest and compare the pros and cons for both approaches!

[Kasper Peulen @KasperPeulen]

Excited about this: got a lightweight async_hooks compatibility shim working in the browser for vitest-plugin-rsc.

This should allow us run Next’s real cookies() and headers() request-context code in browser-mode RSC tests, without mocking those APIs.

@kettanaito

[Kasper Peulen @KasperPeulen]

@thdxr I agree, but how do you get that while:
- being able to still colocate code with test
- keep the ability to statically decide which tests to needs to run
- be fast enough to get feedback back in seconds
- have code coverage of all paths without blowing up your e2e suite

[DHH @dhh]

I've been driving GPT5.5 on low reasoning for the last week+ and it's very good, very efficient. Haven't been tempted to reach for Opus at all. And it's more succinct than Kimi too. Huge leap forward for @OpenAI 👌