Tessera
Folgen

Mathias Lafeldt

@mathias.blog

Indie product engineer obsessed with dev tools, performance & AI. ⚡️ 🔗 mathias.blog

969 Posts705 Follower128 Folgt
Mathias Lafeldt@mathias.blog · last wk.

While Opus 5 rejects any work involving a DuckDB sig bypass I found in February, it's happy to support my game hacking efforts, digging into ROMs using Capstone and casually writing MIPS interpreters to compare results. 🤩

Bild
Hacker News 50@betterhn50.e-work.xyz · last mo.

Claude Design System Prompt https://github.com/Trystan-SA/claude-design-system-prompt (https://news.ycombinator.com/item?id=48792399)

GitHub - Trystan-SA/claude-design-system-prompt: Reverse-engineered system prompt and skill library that turns an LLM into an opinionated, accessibility-aware, AI-slop-resistant design collaborator.

Reverse-engineered system prompt and skill library that turns an LLM into an opinionated, accessibility-aware, AI-slop-resistant design collaborator. - Trystan-SA/claude-design-system-prompt

github.com

Mathias Lafeldt@mathias.blog · last mo.

Bummer that Codex doesn't support disable-model-invocation. Instead, you have to add an allow_implicit_invocation policy to agents/openai.yaml everywhere. I wonder how many skills are subtly "broken" due to this. github.com/mattpocock/s...

Add Codex OpenAI metadata to manual skills · Issue #163 · mattpocock/skills

Hi! Small compatibility suggestion for Codex users. OpenAI Codex skills support optional agents/openai.yaml metadata for invocation policy: https://developers.openai.com/codex/skills#optional-metad...

github.com

Mathias Lafeldt@mathias.blog · last mo.

If a coding agent has fixed the same area 3+ times and review still finds problems, the architecture is wrong, not the code. Stop patching and redesign. I call this agent/review whack-a-mole: every fix you make pops up a new problem somewhere else.

Mathias Lafeldt@mathias.blog · last mo.

It's crazy to expect agents to get it right the first time without any grilling beforehand or adversarial code reviews afterward. Review is also where you enforce your standards. With fresh eyes/context, not in the dumb zone.

Mathias Lafeldt@mathias.blog · 2mo ago

Agent memory is such a bad idea. Way too magic/opaque/unsharable. Embrace AGENTS.md and docs/adr for working instructions and design decisions. Disable and reset agent memory in Claude Code and Codex.

Mathias Lafeldt@mathias.blog · 2mo ago

One of my most useful skills is /squash-commits. It does what you'd expect, but also proposes different commit groupings and is hunk-aware (splitting commits when it makes sense). I use it all the time to clean up branches and PRs. 💪

Haken dran - Das Social-Media-Update@hakendran.org · 2mo ago

SpaceX geht an die Börse, die US-Regierung spricht Exportverbote für Antropic-Modelle aus und mehrere Journalisten und Politiker sollen KI-generierte ungekennzeichnet veröffentlicht haben. Gut, dass wir @pip.net haben, um das mal alles einzuordnen. Heute bei Haken dran:

Finanznihilismus und Fußvolk (mit Pip Klöckner)

Man kann wieder Twitteraktien kaufen! Es gibt dafür allerdings ein paar Wenns und Abers - und man kauft leider auch das ganze Raketenbusiness von Elon Musk mit. Wir nehmen den Börsengang von SpaceX au...

hakendran.podigee.io