Note di Matteo


#claude

L'approccio anti-developer di Anthropic che il feedback degli utenti su Claude Code è imbarazzante:

Boh.

#400 /
23 marzo 2026
/
12:04
/ #anthropic#claude


C'è un bug in Claude Desktop legato agli scheduled task (nuova feature appena lanciata) che manda il codice in loop infinito:

Users with scheduled tasks in Claude Cowork or Claude Code who are in a timezone that observed daylight saving time last night were affected by an infinite loop. When the app tried to locate tasks scheduled during the “skipped” hour, it couldn’t resolve them and got stuck.

Ho l'istinto di dire che è proprio un bug da vibe coding, ma in realtà lavorare con le date e i fusi orari è fonte di numerosi bug da quando la programmazione esiste.

#372 /
8 marzo 2026
/
23:37
/ #ai#claude

Questa mi ha fatto ridere: prompt injection tramite un issue GitHub preso in carico da un workflow GitHub Actions che esegue Claude senza limiti di permessi:

Cline’s (now removed) issue triage workflow ran on the issues event and configured the claude-code action with allowed_non_write_users: "*", meaning anyone with a GitHub account can trigger it simply by opening an issue. Combined with --allowedTools "Bash,Read,Write,Edit,Glob,Grep,WebFetch,WebSearch", this gave Claude arbitrary code execution within default-branch workflow.

#367 /
6 marzo 2026
/
09:30
/ #ai#claude

Someone has to prompt the Claudes, talk to customers, coordinate with other teams, decide what to build next. Engineering is changing and great engineers are more important than ever.

Boris Cherny, creatore di Claude Code

#365 /
4 marzo 2026
/
09:28
/ #ai#claude

Benchmark super interessante che misura le scelte di stack tecnico dei modelli Anthropic (via Claude Code):

The big finding: Claude Code builds, not buys. Custom/DIY is the most common single label extracted, appearing in 12 of 20 categories (though it spans categories while individual tools are category-specific). When asked “add feature flags,” it builds a config system with env vars and percentage-based rollout instead of recommending LaunchDarkly. When asked “add auth” in Python, it writes JWT + bcrypt from scratch. When it does pick a tool, it picks decisively: GitHub Actions 94%, Stripe 91%, shadcn/ui 90%.

Tutti i dettagli nel report completo.

#359 /
27 febbraio 2026
/
14:12
/ #ai#anthropic#claude


"I was wrong"

Meglio di "You're absolutely right", probabilmente.

(Claude Code)

#141 /
11 novembre 2025
/
16:21
/ #ai#anthropic#claude