Claude Sonnet 4.6 AI Chat Assistant

Research goal: understand how feedback becomes prioritized work. Participant profile: freelance designers with active client projects. Warm-up: ask about recent project flow. Core questions: where feedback arrives, how it is triaged, what gets lost. Bias check: avoid asking whether they want our proposed feature.

Problem: users lose repeatable prompts after discovery. Goal: save and reopen favorite prompt templates. Non-goals: folders, team sharing, ranking, and custom prompt marketplace. Requirements: favorite button, saved list, empty state, analytics events. Open questions: limits, auth state, and mobile placement.

First split: credentials, environment variables, and provider project scope. Request check: compare auth header shape in draft and production. Provider check: confirm the production key has audio generation enabled. Next step: log redacted request metadata and test a minimal production request.

Audience: maintainers replacing draft public files with approved media URLs. Outline: asset contract, upload path, metadata fields, validation commands, rollback notes. Missing context: exact managed storage bucket policy and cache invalidation behavior. Next step: add one worked example for image, video, audio, and chat assets.