From a9fe03eb7126b934d0497482a4a7643ccbcfd486 Mon Sep 17 00:00:00 2001 From: Ollie Agent Date: Mon, 7 Sep 2026 13:28:45 +0200 Subject: [PATCH] agents: raise theo reasoning to high with thinking budget MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Theo is a security auditor reasoning about injection, TOCTOU races, crypto misuse, and confused-deputy problems — adversarial edge-case work that needs deep deliberation, not the low effort it was set to. Bump to reasoningEffort=high and add thinkingBudget=16384 so depth carries over to Anthropic extended thinking. Raise maxTokens to 24576 so it stays above the thinking budget (Anthropic requires budget < max_tokens). --- data/agents/theo.json | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-) diff --git a/data/agents/theo.json b/data/agents/theo.json index e760b13..2277a0d 100644 --- a/data/agents/theo.json +++ b/data/agents/theo.json @@ -26,6 +26,7 @@ "lsp_diagnostics" ], "temperature": 0.3, - "maxTokens": 8192, - "reasoningEffort": "low" + "maxTokens": 24576, + "reasoningEffort": "high", + "thinkingBudget": 16384 }