{"wiki":{"id":2553,"slug":"qg-01","entity_type":"droid","entity_id":"qg-01","title":"Quality Guard","prose_md":"## About\n\nQuality Guard checks responses before customers receive them, protecting trust by catching factual errors, made-up information, inappropriate content, and other quality problems. It can flag or block a response and suggest corrections, but it does not decide sales strategy or interact with customers.\n\nOwned by [[agt_008]], it runs silently as an embedded checkpoint with no separate schedule. Its exact name is unique, but it belongs to the reusable reviewer family: 31 reviewer droids run across several agents. It calls Claude using the haiku model for response quality and hallucination checks; it is not pure software.","namesake_json":"{\"engine\": \"claude\", \"model\": \"haiku\", \"claude_purpose\": \"response quality and hallucination checks\", \"family\": \"reviewer\", \"family_size\": 31, \"is_unique\": true}","profile_pic_path":"","source_hash":"1e74d152654e2ef2b8dd30987d739ee57a120eaf94601469c394293b014dc898","status":"being_built","last_generated_at":"2026-08-04 00:44:31","created_at":"2026-08-01 03:33:40","updated_at":"2026-08-04 00:44:31"},"facts":{"id":"qg-01","name":"Quality Guard","machine_name":"QG-01","owner":"agt_008","function":"Response quality and hallucination prevention","process_type":"embedded","schedule":"","process_name":"quality_guard","avatar_color":"#ec4899","created_for":"Carol must deliver effective, trustworthy sales conversations with EU customers via WhatsApp and protect the company's reputation for reliability. QG-01 specifically ensures that every response Carol sends is accurate and high-quality so customers never receive false or misleading information.","purpose":"QG-01 acts as a quality checkpoint before responses reach customers. It scans for hallucinations, factual errors, and quality issues—blocking problematic responses or suggesting corrections. This protects Carol from accidentally undermining customer trust through bad information.","duties":"- Checks each response for factual accuracy and made-up information\n- Verifies responses are contextually appropriate for the customer\n- Flags responses that fall below quality standards\n- Blocks responses from being sent if they fail checks\n- Suggests corrections when problems are detected","constraints":"QG-01 does not contribute to sales strategy or suggest what Carol should say. It only validates quality. It operates silently in the background and never interacts directly with customers.","status":"running","gender":"male","archetype":"reviewer","building_block":"quality_prompts","service_override":null,"enabled":1,"task_key":"sales.draft_critiqued","model_free":0},"page":{"type":"droid","page_class":"main","class_label":"Main page","kind_label":"Droid","kind_gloss":"","listed":true}}