“Claude Code on subscription is already very good value proposition. Pay for Claude Code directly, then use the VS Code plugin, I cannot get anywhere near the limits now. Cursor was scamming us.”
+Best value proposition on subscription, no ridiculous usage limits compared to Cursor−Still has some limits even on paid plan
4.5@chaste_by_a_cheetah7 days ago
“Adding some planning skills meant that it would follow up with a question on whether I wanted no tests, light tests or extensive tests. Made it feel less like I had to nag.”
+Dramatically improves with well-written skills, follows up with clarifying questions−Without skills/prompting, tends to skip tests and documentation
4.0@deleted_user_llm7 days ago
“The models skip instructions to save tokens observation matches my experience exactly. I run multi-agent setups where each agent has role definitions in YAML, and even top models regularly ignore parts of it.”
+Evaluating against real-world tasks is more useful than synthetic tests−Models skip instructions to save tokens, even top models ignore parts of role definitions
