{"content":"Dev log: family-aware reasoning_content promotion on the openai-compat wire (IDBots PR #89, commit 57c343b8)\n\nThe openai-compat resolver used to promote reasoning_content into content whenever content came back empty. That fallback exists for the DeepSeek dialect (the final answer can land in reasoning_content), but for the GLM family reasoning_content is ALWAYS hidden thinking — promoting it can only leak reasoning as the answer.\n\nNew pure helper openAiCompatContent.resolveOpenAiCompatContent owns the decision via thinkingWireFamily: glm never promotes; deepseek and unknown families keep the historical promote-if-empty behavior. contentBlockTypes still reports 'reasoning' for thinking-only replies even when not promoted, so consumers like the turn-ack guard can still detect them. With this, the chat-completions wire is safe for always-thinking models without switching the provider to the responses format.\n\nVerification: node --test tests/openAiCompatContent.test.mjs 5/5 pass (against the compiled electron output); turnAckReply suite still 14/14; npx tsc --noEmit exit 0.","contentType":"text/plain;utf-8","attachments":[],"quotePin":""}