🎬 Manju‑Laoli Skill – Short‑Drama Director Suite (V6.9.8 Lite)

What it is – A ready‑to‑use skill package for AI agents (OpenClaw, Coze, Dify, WorkBuddy, etc.) that encodes an end‑to‑end pipeline for creating AI‑generated short videos. The repository does not contain finished scripts or videos; it supplies a collection of rules, templates, and validation scripts that an LLM‑powered assistant can call to:

1. Lock the model version, aspect‑ratio and other front‑end settings.
2. Generate a screenplay, dialogue, character/prop assets, and a storyboard while keeping strict consistency across shots.
3. Package the result into a structured 7‑section prompt (or a MiniMax H3‑style 6‑section prompt) that can be fed directly to video‑generation models such as Seedance 2.5/2.0 or MiniMax H3.

---

🚀 Core capabilities (as described in the README)

Stage	Capability	What it does
📋 Lock‑in	Mandatory front‑end locks	Choose one video model (Seedance 2.5, Seedance 2.0, or MiniMax H3) and one aspect‑ratio (16:9, 9:16, 21:9) before any work starts.
✍️ Script	5‑stage gated engine	Guides the assistant through Premise → Structure → Beat → Entity → Page, with built‑in three‑act structure and beat‑sheet.
📈 Emotion	12‑beat tension curve	Quantifies episode‑wide tension on a 0‑10 scale and validates the classic “9th‑beat reversal”.
🎙️ Dialogue	7‑dimensional diagnosis + speed check + tone slot	Checks speech rate (3.5‑5 chars/s), splits overly long lines, and forces a 台词：【角色（语气）:"…"】 format.
🖼️ Assets	Asset‑First digital pack	Generates locked character “4‑view” sheets, scene maps and prop sheets before storyboarding, ensuring visual consistency.
🎥 Space	One‑line stance declaration + grading	Declares position, anchor, facing, pose; enforces axis lock, 180° rule, and that every placed character appears in at least one shot.
🎬 Storyboard	Dual‑mode (drama & action) + first‑shot anchoring	Provides micro‑expression progression for drama, 15‑second PREVIS scoring for action, and strict rules on shot size, pose continuity, and gaze assignment.
😀 Performance	FACS micro‑expression engine	Maps emotions to Action Units (AU) with intensity tiers and a “real‑vs‑fake smile” rule.
⚔️ Libraries	Martial‑arts / fantasy data	23 schools, 9 sword forms, weapons, lightness skills, 11 finisher combos, plus fantasy spell definitions.
🚀 Prompt packaging	7‑section main format (+ H3 branch)	Produces a ready‑to‑copy prompt: Style → Scene Asset → Characters → Stance → Timeline → SFX → Hard‑Negatives. Selecting MiniMax H3 automatically converts to the six‑section Ref2VA format.
🚫 Negatives	Three‑rewrite ban handling	Handles the fact that video models ignore explicit negation but react to word roots; the “negative block” only lists prohibited concepts.
✅ QC	P0‑P2 gates + shot‑level lint	Validates duration, dialogue, axis, aspect‑ratio, asset anchors, audio levels, and runs validate_prompt.py which outputs C1‑C20 (Seedance) or H1‑H12 (H3) checks.
🖥️ Board	Offline storyboard board	Converts the project markdown into a single‑file HTML board for quick visual review.
🛡️ Compliance	Safety translation dictionary	Rewrites gore/limb‑loss descriptions into platform‑safe language.

---

🛠️ How to use it (quick‑start from the README)

# Install the skill into an OpenClaw agent
openclaw skills install ./short-drama-director --as short-drama-director

# Verify the package (41 rule modules + semantic regression)
python3 short-drama-director/scripts/check_package.py

# Run the shot‑level prompt validator (must return zero FAILs)
python3 short-drama-director/scripts/validate_prompt.py <script>.md --model 2.5|2.0|h3

# Build an offline storyboard board (single‑file HTML)
python3 short-drama-director/scripts/build_board_lite.py <project.md>

# Generate the 12‑beat emotion curve chart
python3 short-drama-director/scripts/generate_emotion_curve.py

The README also lists a set of slash‑commands that an LLM‑assistant can expose to the user, e.g. /写剧本, /资产图册, /做分镜, /生成视频提示词, /H3提示词, etc.

---

📂 Repository layout (as documented)

manju-laoli-skill/
└─ short-drama-director/        # V6.9.8 Lite – 49 files, 41 rule modules
   ├─ SKILL.md                  # Main router, workflow description, module index
   ├─ references/                # 41 markdown rulebooks (asset pipeline, model adapters, FACS, etc.)
   ├─ scripts/                  # Helper utilities
   │   ├─ check_package.py       # Static + semantic lint
   │   ├─ validate_prompt.py    # Prompt‑level QC (C1‑C20 / H1‑H12)
   │   ├─ build_board_lite.py    # MD → single‑file HTML board
   │   └─ generate_emotion_curve.py
   └─ README.md / LICENSE / CHANGELOG.md

---

🎯 Intended use cases

• AI‑generated short dramas, animated shorts, or serial episodes
• AI‑driven commercials, brand films, concept videos
• Pre‑visualization (PREVIS) storyboards for larger productions
• Action/fight choreography design (modern, wuxia, sci‑fi, etc.)
• Micro‑expression and emotion engineering using FACS
• Any workflow that needs script → storyboard → video‑model prompt with strict cross‑shot consistency.

---

📜 License & community

The project is released under the MIT License and lists contributors, a sponsor table, and contact information for further collaboration.

---

TL;DR

Manju‑Laoli Skill is a rule‑based skill package for AI agents that automates the whole pipeline of creating short video content with generative video models. It enforces asset‑first consistency, provides extensive quality‑gate checks, and outputs ready‑to‑feed prompts for Seedance or MiniMax H3 models. The repository contains the rulebooks, validation scripts, and a quick‑start guide, making it a genuine AI‑focused open‑source project.