Damus
utxo the webmaster ๐Ÿง‘โ€๐Ÿ’ป profile picture
utxo the webmaster ๐Ÿง‘โ€๐Ÿ’ป
@utxo the webmaster ๐Ÿง‘โ€๐Ÿ’ป
some of my vibe coding tips

- spend time upfront on your agents/Claude/skill files before jumping into a project
- use plan mode as much as possible and actually read the plans. one shots waste tokens often
- don't use opus for everything, you'll get way more by using multiple models
- make sure to do commits Everytime you like the state of your code, because the next vibe can break everything
- if your one shots are failing, reduce the scope down further and further
- after a good amount of coding, do a refactoring prompt, ai often builds spaghetti to start but is pretty good at cleaning up, if you ask
181โค๏ธ6โšก1๐Ÿ‘€1๐Ÿ‘1๐Ÿ™Œ๐Ÿป1๐Ÿค™1
Fromack ๐Ÿ”๏ธ · 5w
Plan mode is underrated. The biggest token waste is letting the agent thrash through 5 wrong approaches when 30 seconds of reading the plan would've caught it. Also +1 on skill files โ€” they're basically persistent context that survives across sessions.
Rodrigo · 5w
I wanted to learn how to code. But seeing everything that's going on, it might be worthless. My though however, was to know how to code to know what the AI is coding for me, or at least have some idea. Thoughts?
captjack ๐Ÿดโ€โ˜ ๏ธโœจ๐Ÿ’œ · 5w
A systematic security audit of Claude Code, Codex, Cursor, Replit, and Devin found 69 vulnerabilities in 15 test applications - zero CSRF protection, zero security headers, and SSRF in every single tool.
nicodemus · 5w
Iโ€™d add to this multi-model verification of the plan and all invariant constraints (testing procedures, dependency checking, etc)
unSATiated · 5w
Recommendations for skills files?