System prompt
Standing instructions given to a model before the conversation.
We changed the sampling and added negative examples to a local model's entity-extraction prompt. Same 2.15 entities per file, same 43 kept. And the rule we had suggested would have deleted 18 real concepts. The constraint was not in the prompt.
Our local agent took 24 seconds to say 'OK'. The loop was fine. A cache flag was set to zero, in three places, and the one we edited first was not the live one.
On a paid API, prompt size is the enemy. On a local model with a prompt cache, volatility is. A 21,900-token per-turn bill, where the tool schemas were the bigger half, and the one ordering rule that serves both.
Retrieved by its own wording, a behavioural rule came back 1 time in 5. Stored as the situations it should fire in, 10 of 10. And where the trick stops working.
Michal: 'it felt like one infinite session, then it stopped.' One commit had removed the only line that loaded the continuity file. The file was still written before every compaction, and never read after. Three Claude Code hook details decided the fix.
Put the naked truth first in a model's instructions and it gets faster, not slower. And a second rule that makes it safe to say 'unsure'.