Viral Short Drama Script Engine

Problem

Short-drama drafting can drift when trend selection, story, episode outlines, and episode scripts are produced in isolation. Manual review often misses structural issues such as a character speaking without entering the scene, missing scene-list consistency, or inconsistent naming across episodes. This skill turns the workflow into an explicit, script-controlled state machine.

How It Works

• Fixed stages: it enforces story → structure → script → review → done, so the model cannot skip stages.
• Script-owned IO and checks: compose returns a variable-injected prompt, save persists output and runs deterministic checks, and rollback can redo one episode or the full outline.
• Human-in-the-loop: each stage output is presented for confirmation, while state/story_state.json remains the single source of truth.
• Anti-hallucination checks: it validates character and scene-list consistency, dialogue participant presence, naming consistency, and ending hooks, blocking pass when blockers are hit.
• Creation path: use fetch-trend to gather trending signals, or set a custom direction via --brief / save-brief, then produce story_core, outline.json, and per-episode scripts.

Boundaries

Use it when the target short-drama track or hot-topic direction is clear and per-episode script delivery is needed. It does not cover post-launch completion rate, data feedback loops, or weight evolution. If review fails, the work should not be marked complete; use rollback to redo the affected episode or return to the structure stage.