nikolasi

5 ms·

1. 1.▲
   by nikolasi 7mo ago

   That's a great way to put it — "cognitive scaffolding" is exactly what it is. And yeah, keeping rules minimal is smart, but at some point the project just needs 25 rules and you can't cut them down without losing somethi
2. 2.▲
   by nikolasi 7mo ago

   Glad it resonated! Yeah repeating instructions every N turns was the old approach — SCAN basically does the same thing but with ~20 tokens instead of the full prompt each time. On drift being "mostly gone" — depends on prompt comp
3. 3.▲
   Every agent framework has the same bug – prompt decay. Here's a fix(gist.github.com)

   1 points by nikolasi 7mo ago|5 comments
4. 4.▲
   by nikolasi 7mo ago

   I run 11 agents in 100K+ token sessions. Same problem everyone has: agents follow rules perfectly at first, then drift until the system prompt might as well not exist. SCAN fix: put questions at the end of each prompt section. Before each t
5. 5.▲
   Multi-agent coordination on Claude Code: 8 production pain points and patterns(gist.github.com)

   1 points by nikolasi 7mo ago|0 comments