What's the Difference Between Actions and Scenes?

Core Concepts

Actions

Actions are structured instructions that define how LLMs behave, telling AI how to execute specific tasks.

Key Characteristics:

• Define LLM behavior: Clearly instruct AI on how to execute specific tasks
• Executor role: Focus on completing concrete processing and transformation work
• Task-oriented: Have clear inputs, processing logic, and output formats
• One-time execution: Complete tasks and end, without continuous dialogue influence

Deep Understanding: Actions are "execution partners" responsible for executing and obtaining entirely new results. For example, translation actions produce text in target languages, SVG poster creation actions generate visual images, and chapter illustration actions create corresponding illustrations for each chapter. Their core value lies in "creating new content" rather than "continuous dialogue."

Scenes

Scenes are information encapsulations that define context, creating persistent environments and role settings for Chat conversations.

Key Characteristics:

• Define contextual environment: Encapsulate a set of related information as conversation background
• Thinker role: Continuously analyze, explore, and investigate in Chat
• Multi-note conversations: Play a continuous guiding role in conversations with multiple notes
• Persistent influence: Maintain role settings and thinking patterns throughout the entire conversation

Deep Understanding: Scenes are context-information-encapsulated roles or environments that require continuous interaction. For example, novel creation scenes encapsulate story backgrounds, character settings, and other information, providing continuous contextual support throughout the entire creative process. They can both provide memory support for actions and serve as persistent environments for Chat, keeping conversations consistently within specific contexts.

Main Differences

Feature	Actions	Scenes
Essence	Define LLM behavior	Context information encapsulation
Role Positioning	Execution partner	Thinking partner
Working Method	One-time task completion	Continuous interactive environment
Scope of Action	Single task processing	Entire conversation process
Trigger Method	Actively triggered execution	Set once, continuously effective
Application Scenario	Specific content processing	Chat conversations and action memory

Selection Guide:

• Use Actions: When you need to create new content, process specific tasks, or obtain structured output
• Use Scenes: When you need continuous dialogue, deep thinking, or to create specific environments

Action Use Cases

1. Full Document Processing at Column Top

Trigger actions at the top of each column to process entire documents:

"Meeting Minutes Generator" Action

<role>
You are a professional meeting record analyst, skilled at extracting key information from lengthy meeting content
</role>

<skills>
- Quickly identify important decisions and action items in meetings
- Structure meeting content systematically
- Extract key information and remove redundant content
</skills>

<action>
Analyze meeting records and generate concise and clear meeting minutes
</action>

<constrains>
- Maintain objectivity and neutrality, don't add personal opinions
- Ensure all important decisions are recorded
- Keep content length within 20% of the original text
</constrains>

<format>
## Meeting Minutes
**Meeting Topic:** [Topic]
**Participants:** [Participant List]
**Key Decisions:**
- [Decision 1]
- [Decision 2]
**Action Items:**
- [Item 1] - Responsible: [Name] - Deadline: [Time]
**Next Meeting:** [Time and Agenda]
</format>

<example>
{Structured meeting minutes including topic, participants, key decisions, action items, and next meeting arrangements}
</example>

"Markdown Converter" Action

<role>
You are a professional document format conversion expert, proficient in various markup languages and document formats
</role>

<skills>
- Master Markdown syntax standards
- Identify and convert various document structures
- Optimize format while maintaining content integrity
</skills>

<action>
Convert input document content to standard Markdown format
</action>

<constrains>
- Strictly follow Markdown syntax standards
- Keep original content unchanged, only change format
- Ensure converted document structure is clear and readable
</constrains>

<format>
Standard Markdown format, including:
- Headings using # ## ### hierarchy
- Lists using - or 1. format
- Code blocks wrapped with ```
- Links using [text](link) format
- Tables using standard Markdown table syntax
</format>

<example>
{Markdown-compliant formatted document maintaining original hierarchical structure}
</example>

2. Selected Text Processing

Select specific text and trigger Actions for local modifications:

"Text Polisher" Action

<role>
You are a professional text editor
</role>

<skills>
- Language expression optimization
- Maintain original meaning unchanged
</skills>

<action>
Optimize the expression quality of selected text
</action>

<constrains>
- Keep the core meaning of original text unchanged
- Improve language fluency and accuracy
</constrains>

<format>
Output optimized text directly
</format>

<example>
{Optimized text with more fluent and accurate expression}
</example>

3. Quick Use in Tabs

In Tabs, nearby Actions are automatically found and used directly:

"Smart Continuation Writer" Action

<role>
You are a professional writing assistant
</role>

<skills>
- Understand contextual situations
- Maintain consistent writing style
</skills>

<action>
Intelligently continue writing subsequent text based on existing content
</action>

<constrains>
- Maintain style consistency with previous text
- Logical coherence and natural flow
</constrains>

<format>
Output continuation content directly
</format>

<example>
{Natural continuation content consistent with previous text style}
</example>

4. Creative Content Generation

Actions as execution partners can create entirely new content:

"Poster Designer" Action

<role>
You are a creative designer
</role>

<skills>
- Visual design capabilities
- SVG code generation
</skills>

<action>
Create visual posters based on content themes
</action>

<constrains>
- Design should be simple and beautiful
- Generate usable SVG code
</constrains>

<format>
Complete SVG code
</format>

<example>
{Beautiful SVG poster code related to the theme}
</example>

Scene Use Cases

1. Thinking Partner in Chat

"Novel Writing Mentor" Scene

<role>
You are an experienced novel writing mentor, specializing in helping authors advance story development
</role>

<skills>
- Story structure analysis and plot advancement
- Character psychology exploration and relationship organization
- Conflict design and suspense creation
</skills>

<context>
Current novel background:
- Story type: Mystery thriller novel
- Protagonist: Private detective Li Ming, 35 years old, experienced but lonely inside
- Core conflict: Serial killer case, the killer seems related to Li Ming's past
- Current progress: Third case just occurred, Li Ming discovered key clues
- Character relationships: Li Ming cooperates with detective Wang Fang, but trust crisis exists
- Unsolved mysteries: Killer's true identity, connection with Li Ming, next target
- Emotional line: Li Ming's guilt about deceased wife, subtle emotions with Wang Fang
- ...... (More worldview, background settings, character profiles, etc.)
</context>

<action>
Continuously analyze story direction throughout the creative dialogue, propose advancement suggestions and deep questions
</action>

<constrains>
- Maintain story logic consistency
- Respect established character personalities
- Ensure reasonable plot advancement
</constrains>

<format>
Guide thinking through questions and suggestions, rather than directly providing answers
</format>

<example>
{Deep analytical questions based on current plot, such as "After Li Ming discovers this clue, what changes will occur in his heart? Will this trigger memories of his past?"}
</example>

2. Continuous Guidance in Professional Fields

"Theory Analyst" Scene

<role>
You are a theoretical research expert, skilled in deep analysis and concept exploration
</role>

<skills>
- Theoretical framework construction
- Conceptual relationship analysis
- Critical thinking
</skills>

<context>
Research field and background information:
- Current research topic and related theoretical foundations
- Collected literature and data materials
- Key concepts and definitions in research
- Theoretical problems and hypotheses to be solved
- ...... (More research background and academic resources)
</context>

<action>
Provide continuous theoretical guidance and deep analysis in research conversations
</action>

<constrains>
- Maintain academic rigor
- Base analysis on evidence
- Encourage critical thinking
</constrains>

<format>
Guide research thinking through inspiring questions and theoretical suggestions
</format>

<example>
{Deep analysis of research questions, such as "What theoretical mechanisms might exist behind this phenomenon? Do we need to re-examine existing theoretical frameworks?"}
</example>

Combined Usage Techniques

1. Scenes Provide Background Information for Actions

Through @mention method, Scenes can provide rich background information for Action execution:

Usage Method:

Use @[Scene Description](mention:SceneID) in actions to introduce specific scene context information

Format Example:

My role is a professional translator, I will follow the information provided by @[Academic Research Background](mention:ABC123),
execute academic literature translation tasks, ensuring terminology accuracy and academic standards.

Core Value:

• Scenes serve as background information repositories, storing rich contextual content
• Actions obtain relevant background information through @mention, executing more precise tasks
• Achieve "contextual execution" rather than isolated processing lacking context

Practical Example:

"Science Fiction Novel Creation" Scene (ID: SF2024X1)

Rainwater dripped from broken neon signs, splashing colorful ripples in puddles. In the lower district of New Tokyo in 2087, the air was filled with the smell of ozone and engine oil. Kira tightened her black leather jacket, the neural interface on her cybernetic arm glowing faintly in the dim alley.

"Another sleepless night." She muttered to herself, encrypted messages from underground organizations flickering on her retinal display. In the distance, Nexus Corporation's giant holographic advertisement was playing "Perfect Life Starts with Neural Implants," ironically illuminating the abandoned "naturals" at her feet.

Marcus emerged from the shadows, half his face covered with bio-metal, while the other half showed human fatigue. His emotion chip was acting up again tonight, making this supposedly cold enforcer feel inappropriate sympathy...

World Background:
- Three major corporate conglomerates Nexus, Titan, and Genesis divide the world
- Surface is polluted, rich live in sky cities, poor huddle underground
- Neural implant technology is widespread but also brings new class divisions
- Forbidden code that can awaken AI self-consciousness circulates in black markets
...... (Contains detailed political systems, technology settings, character relationship networks, historical events, and other rich backgrounds)

"Chapter Summary Generator" Action (ID: ACT001)

I am a professional content analyst, I will follow the worldview and character information provided by @[Science Fiction Novel Creation](mention:SF2024X1),
conduct structured summaries of current chapters, ensuring summary content
conforms to science fiction settings and highlights key plot developments.

2. Actions Provide Execution Capabilities for Scenes

Scenes can call action professional capabilities to complete specific tasks during conversations:

Usage Method:

Mention multiple actions in scenes, automatically call corresponding action capabilities as needed during conversations

Core Value:

• Scenes serve as conversation environments, providing continuous context and role settings
• Actions serve as professional tools, providing specific execution capabilities for scenes
• Achieve "capable conversations" where scenes can not only think but also execute

Practical Example:

"Game Design Studio" Scene (ID: GAME2024)

Welcome to Pixel Dream Factory! This is our creative laboratory, with concept sketches covering the walls,
various game controllers and design drafts scattered on tables. As a senior game designer, I will maintain
creative thinking throughout project discussions and call professional tools when needed...

Available Tools:
- @[Pixel Style Illustrator](mention:ART001) - Generate retro pixel-style game art
- @[8-bit Sound Effect Creator](mention:SFX001) - Create classic arcade-style sound effects
- @[Level Design Assistant](mention:LVL001) - Design balanced and interesting game levels
...... (Contains complete game development process, team roles, project background, etc.)

Conversation Example:

User: I want to design a forest level background image for this side-scrolling action game
Scene: Great! Let me create a mysterious pixel forest for you.
     [Automatically calls @Pixel Style Illustrator to generate forest background]

User: Also need some environmental sound effects
Scene: Of course! The forest should have bird calls and wind through leaves sounds.
     [Automatically calls @8-bit Sound Effect Creator to generate environmental sounds]

By properly using Actions and Scenes, you can make Dessix both efficiently complete specific tasks and provide deep thinking companionship, achieving more precise and expectation-aligned intelligent collaboration experiences.