[New] mobieusAI: Text-only input — no image, screenshot, or voice input for the assistant or helpdesk

Area: mobieusAI (audit p11) · Surface: AnthropicClient::complete (text-only) + all AI entry points · Dimension: competitor-gap · Severity: minor

Intercom Fin and Zendesk let an end user attach a screenshot to a ticket and have the AI read it; community AIs increasingly accept image context. mobieusHelp tickets routinely include screenshots, but the reply-suggest/summary/categorize hooks see only text, ignoring the most informative part of many support tickets. Claude supports image input natively, so this is an unused capability of the model we already call.

Evidence

platform/src/Services/AI/AnthropicClient.php:95-98 — message content is hardcoded to a single text block: `'content' => [['type' => 'text', 'text' => $userPrompt]]`. A grep for `image|vision|base64.*media|audio|voice` in AnthropicClient and the CM InferenceClient returns nothing. Every controller passes plain `$_POST` strings (e.g. AssistantAIController.php:28 reads `question` as a string).

Suggested fix. Allow image content blocks in AnthropicClient::complete (accept an optional attachments array of {type:image, source}). Wire helpdesk ticket attachments and the assistant compose box to pass images through, gated by the existing per-feature caps.

Filed by the automated tenant-app audit and adversarially evidence-verified. Status: verified. Open — not yet actioned.