fix(ui+search): approachable starter prompts, animation timing, search robustness

fix(agent): remove content moderation from auditor

fix(agent): deterministic audit, synthesizer grounding, structured output reliability

fix(agent): thinker prompt injects real query, auditor nonsense-check tightened

fix(agent): HITL resume via Command, cap reasoning, auditor nonsense filter, real web search

fix: simplify tools to web_search only and break auditor rejection loop

fix: enhance auditor's JSON parsing and error handling

fix: normalize auditor policy feedback

fix: improve auditor JSON handling and worker config

fix: use null resume for HITL stream

chore: lower max_new_tokens for SmolLM

Deploy: DeepTrust workspace (v2 clean history)