✳flâneur — a map of the web's best reading
Softerware - by Mike Caulfield - The End(s) of Argument
mikecaulfield.substack.com · 661 words · saved by 1 readers
Keep your Claude skills loose for better performance
Saw this earlier today, and it’s funny because it’s true, as they say: I've been doing some large scale things with Claude Code and Claude Cowork recently, and the thing I will tell you about these new “skills” based prompting frameworks is they are best developed as a set of values/objectives/reminders and a bit of code for those things that are tiresome to watch the LLM fumble toward each time. Any more than that? You’re probably mucking things up. These architectures are what I call "softerware", and the trick is to get some of the predictability of software while avoiding the rigidness…
Explore this link on the map →related reading
- A Guide to Claude Code 2.0 and getting better at using coding agents – sankalp's blogsankalp.bearblog.dev
- How I Use Claude Code | Philipp Spiessspiess.dev
- Best practices for Claude Code - Claude Code Docsanthropic.com
- Malleable software in the age of LLMsgeoffreylitt.com
- Don’t Outsource Your Thinkingteltam.github.io
- How We Use Claude Code Skills to Run 1,000+ ML Experiments a Dayhuggingface.co
- GitHub - multica-ai/andrej-karpathy-skills: A single CLAUDE.md file to improve Claude Code behavior, derived from Andrej Karpathy's observations on LLM coding pitfalls. · GitHubgithub.com
- My Claude Code Psychosis - by Jasmine Sunjasmi.news
- How Anthropic teams use Claude Code | Claudeanthropic.com
- Claude Code Cheat Sheetcc.storyfox.cz
- How Claude Code is built - by Gergely Orosznewsletter.pragmaticengineer.com
- Building personal software with Claude - Made of Bugsblog.nelhage.com