Journal
Things we learn
What we learn while building, including the details that rarely get written down.
Anti-Slop’s “after” takes more than 40 rules: it also needs design direction
A source-based review of Anti-Slop: its 40 rules guide agents, the “after” example also uses DESIGN.md, and its code skill only edits comments. With three original project images and corrected Claude Code installation instructions.
The Instagram Agent Skill humanizer deletes the Persian ZWNJ, and your emoji joiners
Instagram Agent Skill is thirteen free, MIT-licensed Claude skills with no signup and no API key. Its humanizer also deletes ZWNJ, ZWJ and direction marks, and detect.py scores correctly written Persian as more machine-like. Read this before you install it.
Claude writes over 80% of Anthropic’s code. Written instructions make the difference
From Anthropic’s own reports: Claude writes over 80% of the code it merges, but the real gain comes from what teams write for it: CLAUDE.md, skills and commands. In data analytics, accuracy never passed 21% without skills and topped 95% with them. And where it still falls short.
Your library is miscounting Persian reading time
Content libraries often count words with /[a-zA-Z]+/. For Persian text, the result is almost zero, with no error to warn you.
Inside preserve-3d, z-index is not enough
Four cards with carefully ordered z-index values, and a 3D shape cutting through them. The problem was depth, not size.
In other languages
Not written in this language yet. Each post opens in the language it was written in.
توسعه‌دهنده‌ها فکر می‌کردند با AI ۲۰٪ سریع‌ترند؛ کرنومتر گفت ۱۹٪ کندتر
در آزمایش METR، توسعه‌دهنده‌های باتجربه با AI ۱۹٪ کندتر کار کردند و بعدش هنوز فکر می‌کردند ۲۰٪ سریع‌تر شده‌اند. یک سال بعد آزمایش تکرار نشد، چون خیلی‌ها دیگر حاضر نبودند بدون AI کار کنند. برای مدیری که بودجهٔ AI را امضا می‌کند، فاصلهٔ حس و اندازه‌گیری مهم‌تر از خود عدد است.