Eight toolbar fixes. Zero effect. One Apple line for the foldable brought Cancel and Save back on the Duo simulator.
Also: faking a fold with simctl crash-looped SpringBoard 15 times in 90 seconds. teapawt.net/oscarg908/status/21184…
Óscar Gallego retweeted
Leyendo decenas de reseñas de apps de mantenimiento, la queja que más se repite no es una función que falta.
Es "la app perdió mi historial": fechas borradas tras una actualización, móvil nuevo y todo a cero.
Si llevas la moto en una app, exporta una copia de vez en cuando.
my first prototype of auto trip tracking for my iOS app scanned for a BLE beacon
it broke whenever iOS suspended the app: background discovery events got batched and delivered in one go, so trips stayed open or ended halfway
replaced it with Shortcuts + App Intents. car Bluetooth or CarPlay connects, trip starts. disconnects, trip ends
the catch: apps can't create those automations for you
anyone found a better way to detect a car trip on iOS without asking the user to set up automations?
my reviewer agent sent back 4 of 5 tasks in one batch. the trick was running it in a clean context on a stronger model and telling it to fail when in doubt. a reviewer that shares the coder's context just agrees with it
tried Vercel BotID on my contact form twice
first time it blocked real visitors
second time it lasted about 17 minutes and blocked me. "Message blocked" from a normal browser
back to a honeypot, a time trap, a link filter and a per-IP firewall limit
what do you use against form spam that doesn't block real people?
expected Astro 7 to be another afternoon like the v6 upgrade
it was pnpm up, a green build and 217 green tests
two things that can still bite: the new Rust compiler stops quietly fixing malformed HTML, and if you use remark or rehype plugins you now install the markdown-remark package yourself
anyone hit either one?
repartir modelos por tarea ayuda con los límites, pero lo que más me cambió fue otra cosa
un agente aparte, en contexto limpio y con un modelo más fuerte, que revisa lo que hizo el otro y tiene que suspender si duda
en una tanda de cinco tareas devolvió cuatro a arreglar
¿alguien más revisa con un modelo distinto al que construye?
el prompt y sus límites: oscargallegoruiz.com/blog/ve…
today I tried giving away a free year of my app to 10 people on Instagram, all they had to do was use it and tell me what's wrong
signups so far: 0
turns out even free is a tough sell lol
has this happened to you too? how did you bounce back
¿qué comprobáis antes de dar por bueno lo que os devuelve un agente?
los tests miran el código
lo que dice del código (el README, la descripción de la PR, las cifras) no lo ejecuta nadie, y ahí es donde suele fallar





