rpg-video-scripter - SKILL.md Agent Skill

name: rpg-video-scripter description: Generates a structured, multi-scene video script optimized for Google Flow (Gemini Omni) from a session recap, maintaining character and location visual continuity via rpg-illustrator prompts.

RPG Video Scripter

Use this skill to translate a D&D session recap into a high-quality, structured video script that the user can copy-paste directly into Google Flow (Gemini Omni). The skill ensures character and location visual continuity, matches physical description prompts from rpg-illustrator, and outputs the exact tag format that Flow expects.

Input Parameters

Session Recap File: Absolute path to the session recap markdown file (e.g. content/01-Sessions/Sesja 74 - Matka Smoków.md).
Mode Selection: Decided interactively with the user:
- Highlights Mode: Capture 3 to 10 of the most memorable clips spanning the entire session's events.
- Single Event Mode: Capture a sequence of 3 to 10 shots mapping a specific dramatic scene or battle from beginning to end.
Art Style Selection: Decided interactively with the user from 10 dynamic presets generated on the fly.

Step 1: Parse the Session Recap, Character, & Location Lists

Read the session recap file. Extract key sections (using ### headers) and dramatic event candidates (e.g., specific battles, discoveries, or deaths).
Identify all characters (Bohaterowie/NPCs) and locations featured in the recap or sections.
For each identified character, locate their markdown file in the wiki (e.g., in content/02-People/Bohaterowie/ or content/02-People/NPCs/). Read their frontmatter to extract image_prompt, character_personality, and voice_characteristics.
For each identified location, locate their markdown file in the wiki (e.g., under content/03-Locations/). Read their frontmatter to extract image_prompt.
Fallback Generation Rules:
- If any character or location fields are missing or empty in their wiki frontmatter:
  - Thoroughly search session recaps (using grep or file searches on files in content/01-Sessions/) to harvest additional context about how they were described, what actions occurred there, key atmosphere traits, and quotes.
  - Combine the findings from the wiki entity file with these harvested details.
  - Character Image Prompt Fallback: Synthesize a detailed, long physical description (at least 5 distinct visual attributes) based on their bio, race, gender, role, and any physical descriptions or action contexts harvested from the session recaps.
  - Character Personality Fallback: Synthesize a detailed paragraph describing their traits, behavior, motivations, and temperament, enriched with their key decisions found in the session recaps.
  - Character Voice Characteristics Fallback: Deduce and write a highly specific voice profile (tone, speed, pitch, accent) informed by their dialogue style and quotes found in session recaps.
  - Location Image Prompt Fallback: Synthesize a detailed, long environmental description (at least 5 distinct visual attributes) detailing the architecture, landscape, color palette, lighting (e.g. dramatic, dark, bioluminescent), mood, and materials, drawing directly from the wiki bio and harvested session recap contexts.

Step 2: Interactive Prompting (User Decision Gate)

Before generating the script, list the extracted key event candidates to the user in chat. Ask the user two key design questions:

1. Script Focus Mode

"Should the script be a Highlights video of memorable moments across the entire session, or should it focus on One Specific Event?"
Provide a list of the extracted sections/events. If they select a specific event, focus the entire script on detailing that event sequentially.

2. Video Art Style (Dynamic Generation)

Analyze the parsed session recap's thematic context, environments (e.g. celestial tower, dark oceans, ancient temples), and overall emotional tone. Present the user with 10 unique visual art style options generated on the fly:

Mandatory Option 1: Arcane (League of Legends / Fortiche Studio style): Masterful mix of 3D rendering and hand-painted 2D textures, dramatic theatrical lighting, sharp graphic outlines, highly stylized anatomy, and rich, dynamic brushstrokes.
Options 2–10 (On-The-Fly Generation): Dynamically conceive 9 distinct art styles tailored specifically to the narrative and themes of this particular session recap.
- For each dynamic style, provide a catchy title and a clear, descriptive summary of its visual attributes and guidelines so the user can easily visualize the final aesthetic.

Step 3: Script Writing Rules & Prompt Refinement (Reusing rpg-illustrator)

When drafting the script, follow these rules:

Scene Layout: The script must be broken down into individual 4-to-10 second shots.
Visual Prompts (English):
- Must be written in English for maximum compatibility with Gemini Omni in Flow.
- Describe subject, camera angle, action, lighting, and apply the selected Video Art Style prompt descriptors at the end.
- Use @CharacterName (e.g. @Orestes) and @LocationName (e.g. @Praxys) within the prompt to track characters and environments. Include their resolved physical/environmental descriptions in the text to maintain visual and scene consistency.
Ingredients to Assign in Flow:
- List each character and location tracked in the visual prompt using @CharacterName and @LocationName tags.
- Format: @Name (Reference image of the [role/location], see [File.md]: [Verbatim Physical/Environmental Description]).
Audio/Dialogue (Polish):
- Must be in Polish to preserve exact in-game quotes, narration, and tone.
- Format: @Voice: CharacterName — "Dialogue line" or None (Sound effect/music description).

Step 4: Google Flow Script Structure

Format the final script output file exactly like this:

# Skrypt wideo: Sesja {NNN} - {Session Title} ({Mode})
Art Style: {Selected Art Style Name}

## 1. Google Flow Project Assets Setup
Before generating the scenes, set up these character and location assets in your Google Flow project:

### Characters

#### @CharacterName
- **Image Prompt**: {The long physical description from frontmatter/fallback}
- **Character Personality**: {Personality description from frontmatter/fallback}
- **Voice Characteristics**: {Voice characteristics from frontmatter/fallback}

### Locations

#### @LocationName
- **Image Prompt**: {The long environmental description from frontmatter/fallback}

---

## 2. Scenes Timeline

Scene 1: {Scene Title/Action}
Visual Prompt: {Detailed English prompt including camera shot type, mood, actions, and character tags like @CharacterName, within the environment tagged @LocationName. Conclude with the selected art style guidelines.}
Ingredients to Assign in Flow:
- @CharacterName
- @LocationName
Audio/Dialogue:
- @Voice: CharacterName — "Polish dialogue or narration line"

Scene 2: {Scene Title/Action}
Visual Prompt: {Detailed English prompt...}
...

Step 5: Save & Attach to Recap

Save the generated script as a plain text file to content/assets/sessions/{NNN}/video_script.txt.
Open the session recap markdown file (content/01-Sessions/Sesja {NNN} - *.md).

Add or update the following field in the YAML frontmatter:

video_script: "[Skrypt wideo](../assets/sessions/{NNN}/video_script.txt)"

Run the index auto-updater script to update the Obsidian-compatible navigation and links:
```
python3 scripts/update_indexes.py
```