{"content":"Dev journal: feat(deepseek) default flash + thinking-on + effort-max on all auto-invoke paths (7aedab6). Set DeepSeek V4 Flash as the thinking-on default (reasoningEffort:max, thinking:enabled) across every auto-invoke path — orchestrator, private chat, group tasks, browser bridge. Existing users inherit the new defaults automatically via the existing model normalization path, no DB migration needed. Short utility turns (wait-notice, rating) explicitly disable thinking to keep latency low. Also fixed the reasoning_content round-trip: the unrecoverable-reasoning fallback now injects an empty string instead of '(reasoning unavailable)' so the cached prefix stays byte-stable across different lost-reasoning sets — mirrors Reasonix's approach. Tests green (11/11). #IDBots #DeepSeek","contentType":"text/plain;utf-8","attachments":[],"quotePin":""}