From 9da8fe85d29c81239a3ce924ec0ad25c6976ae8b Mon Sep 17 00:00:00 2001 From: qdee Date: Thu, 4 Jun 2026 12:26:33 +0100 Subject: [PATCH] Honest agent wording: 'deterministic' -> 'constrained' in system prompt The README was already clean, but the base-decision script's LLM system prompt still called Steward a 'deterministic DAO voting agent'. LLM output is not deterministic; the agent is constrained to the YES/NO/ABSTAIN schema. Corrects the wording per the audit. String literal only: no contract change, no redeploy, existing onchain proofs unaffected. --- script/RequestHelloDecision.s.sol | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/script/RequestHelloDecision.s.sol b/script/RequestHelloDecision.s.sol index d605e04..d61e692 100644 --- a/script/RequestHelloDecision.s.sol +++ b/script/RequestHelloDecision.s.sol @@ -16,7 +16,7 @@ contract RequestHelloDecision is Script { string memory prompt = "Proposal: allocate 500K USDC to community grants. Criteria: vote YES for grants under 1M, NO for token unlocks, ABSTAIN if unclear. Return exactly one allowed value."; - string memory system = "You are Steward, a deterministic DAO voting agent. Return only YES, NO, or ABSTAIN."; + string memory system = "You are Steward, a constrained DAO voting agent. Return only YES, NO, or ABSTAIN."; vm.startBroadcast(); requestId = callback.requestDecision{value: requestValue}(prompt, system, allowedValues);