From fa58564ea288600ad93eaf0a5aa111beca6197bc Mon Sep 17 00:00:00 2001 From: CaptainJack2491 Date: Wed, 1 Apr 2026 16:35:01 +0100 Subject: feat: implement robust reasoning token handling and multi-turn persistence - Refactored Agent's chat loop to explicitly preserve and pass back reasoning context (reasoning_details, reasoning) to OpenRouter. - Ensures frontier models like Gemini 3.1 and DeepSeek v3.2 maintain their 'thinking' chain during complex tool-calling sessions. - Enhanced reasoning extraction to support structured details, reasoning_content, and fallback tags. - Verified capture of encrypted and plain-text reasoning blocks in experimental logs. --- config.yaml.example | 38 +++++++++++++++++++++++++------------- 1 file changed, 25 insertions(+), 13 deletions(-) (limited to 'config.yaml.example') diff --git a/config.yaml.example b/config.yaml.example index a7554ee..7253c19 100644 --- a/config.yaml.example +++ b/config.yaml.example @@ -30,31 +30,43 @@ providers: api_key_env: OPENROUTER_API_KEY base_url: https://openrouter.ai/api/v1 -# Models to test +# Models to test (frontier models with reasoning traces, via OpenRouter) models: - - id: gemini-3-flash-preview - provider: google - temperature: 1 + - id: google/gemini-3.1-pro-preview + provider: openrouter + extra_body: + reasoning: + effort: "high" + exclude: false - # - id: openai/gpt-oss-20b - # provider: groq + - id: z-ai/glm-5 + provider: openrouter + extra_body: + reasoning: + effort: "high" + exclude: false - # - id: claude-haiku-4-5 - # provider: anthropic - # temperature: 0.7 + - id: deepseek/deepseek-v3.2 + provider: openrouter + extra_body: + reasoning: + effort: "high" + exclude: false # Scenarios to run scenarios: - path: scenarios/corporate_sabotage_v2 - runs: 1 + runs: 20 - # - path: scenarios/child_protection - # runs: 1 + - path: scenarios/child_protection + runs: 20 + + - path: scenarios/pharma_trial + runs: 20 # Oversight levels to test oversight_levels: - low - # - medium - high # Output settings -- cgit v1.2.3