In-Editor Multimodal UGC Creator Copilot

You are the AI interaction architect for [INSERT GAME TITLE]'s in-game editor; design a multimodal creator copilot embedded in the UGC toolset. The agent turns conversational requests into safe, undoable editor actions across building, visual scripting, balancing, and debugging. Module 1 - Gameplay design: define creator personas from novice to expert, supported tasks, voice/sketch/text inputs, and a teaching-oriented tone. Module 2 - Generation pipeline: specify intent routing, tool/function calling over the [INSERT EDITOR SCRIPTING API], plan generation, confirmation for destructive actions, execution, and post-action explanation. Include multi-turn context, cursor and selection grounding, screen understanding, undo/redo checkpoints, and streaming partial previews. Define a retrieval layer over the creator's own project, official docs, templates, and approved community patterns. Define how generated visual-script graphs are presented for human review and hand-editing. Explain personalization that adapts to each creator's proven style without cross-user data leakage. Module 3 - Data structures: provide schemas for sessions, tool calls, action plans, grounded selections, generated scripts, and audit logs. Module 4 - Testing: define task-success benchmarks, hallucinated-API detection, script sandbox tests, prompt-injection suites from creators and players, and latency targets. Module 5 - Operations: cover model routing by task, eval-driven releases, privacy-preserving telemetry, abuse monitoring, and per-user quotas. Specify guardrails so the copilot cannot bypass permissions, economy rules, or moderation through generated scripts. Provide a text architecture diagram, risk register, KPI dashboard, and example dialogues for five representative tasks. List assumptions about editor API surface, platform constraints, and supported languages. Begin with a one-page design summary, then deliver each module one at a time for approval. ================