Prompt Templates

Prerequisites: Before reading this guide, we recommend familiarizing yourself with the system prompt configuration in Bot Management.

Monstrum’s prompt template system lets you customize the prompts a Bot uses in different scenarios, with support for workspace-level and Bot-level overrides.

Overview

When a Bot processes each message, the platform automatically assembles a complete system prompt. This prompt is not a single fixed block of text — it is composed of multiple layers of content stitched together. Prompt templates control the foundational layer: the Bot’s core behavioral instructions.

Prompts are resolved through a three-tier priority chain:

Bot-level override  >  Workspace-level template  >  System default

If a prompt is set at the Bot level, the Bot’s version is used
Otherwise, the workspace-level version is used
If the workspace has not set one either, the system default is used

9 Prompt Keys

The platform manages 9 prompt templates covering all aspects of Bot operation:

Core Prompts

Key	Purpose	Supported Variables
`default_task_system`	System prompt for task mode	`{bot_name}`, `{bot_description}`
`default_session_system`	System prompt for session mode	`{bot_name}`, `{bot_description}`

These two are the most critical prompts, defining the Bot’s fundamental behavior in tasks and sessions.

Scenario Prompts

Key	Purpose
`group_chat`	Additional prompt for group chats (injected only in session-mode group chat scenarios)
`planning`	Instructions for planning reasoning mode
`adaptive`	Instructions for adaptive reasoning mode

Internal Prompts

Key	Purpose	Supported Variables
`memory_extraction`	Memory extraction prompt	`{current_section}`, `{conversation}`
`memory_extraction_system`	System role for memory extraction	—
`conversation_summary`	Conversation compression prompt	`{conversation_text}`
`conversation_summary_system`	System role for conversation compression	—

Internal prompts are used by the platform’s automated features (memory extraction, conversation compression) and typically do not need modification.

System Prompt Assembly Order

The final system prompt used by a Bot is composed of the following parts in order:

Session Mode

Base template: The Bot’s custom system prompt (or the default_session_system template)
Group chat instructions: The group_chat prompt (appended only in group chats)
Available resources: ## Available Resources — auto-generated summary of resources and tools
Bot memories: ## Bot Memories — memory content organized by scope
Skill instructions: ## Skills — content from enabled Skills

Task Mode

Base template: The Bot’s custom system prompt (or the default_task_system template)
Available resources: ## Available Resources
Bot memories: ## Bot Memories
Reasoning mode: planning or adaptive instructions (if a reasoning mode is configured)
Skill instructions: ## Skills

The “System Prompt” field in Bot settings replaces the base template in step 1. However, the content in steps 2-5 is always appended automatically and is not affected.

Editing Workspace-Level Prompts

Click Prompt Templates in the left navigation bar
The page lists all 9 prompt keys
Each key shows its current state: System Default or Customized
Click the edit button to modify the prompt content
Save

Modifying a workspace-level prompt affects all Bots that use the default value (those without a Bot-level override).

Reset to Default

Click the Reset to Default button to restore the workspace-level prompt to the system default.

Editing Bot-Level Prompts

Navigate to the Bot detail page and select the Prompts tab
The page lists all 9 prompt keys
Each key shows its current source: System Default / Workspace / Bot Custom
Click the edit button to modify the prompt content
Save

Bot-level prompts have the highest priority. Once set, the Bot will no longer use the workspace or system default version.

Reset to Workspace Default

Click the Reset to Workspace Default button to clear the Bot-level override and fall back to the workspace-level or system default value.

Variable Support

Some prompt templates support variable substitution:

Variable	Available In	Description
`{bot_name}`	Task/session system prompts	The Bot’s name
`{bot_description}`	Task/session system prompts	The Bot’s description
`{current_section}`	Memory extraction prompt	Current memory content
`{conversation}`	Memory extraction prompt	Conversation history
`{conversation_text}`	Conversation compression prompt	Conversation text to be compressed

Common Use Cases

Standardizing Bot Style

Set default_session_system at the workspace level so all Bots use a consistent response style:

You are {bot_name}, {bot_description}.

Please reply in Chinese and maintain a professional, concise style. Avoid using emojis in your responses.

Improving Memory Extraction Quality

Customize the memory_extraction prompt to fine-tune the memory extraction rules:

Extract important information from the following conversation. Focus on:
1. Preferences explicitly expressed by the user
2. Technical decisions related to the project
3. Conventions that need to be remembered long-term

Current memories:
{current_section}

Conversation content:
{conversation}

Customizing Group Chat Behavior

Modify the group_chat prompt to control Bot behavior in group chats:

You are in a group chat. Please note:
- Only respond to messages relevant to you
- Do not repeat what others have already said
- Keep your responses brief

FAQ

Bot behavior hasn’t changed after modifying prompts

Existing active sessions use the prompts from when they were created. Start a new conversation to use the updated prompts
Check whether you modified the workspace-level or Bot-level prompt. If the Bot already has a custom prompt, modifying the workspace-level prompt will have no effect

Not sure which prompt is currently in use

In the Bot detail page under the Prompts tab, each key is labeled with its source (System Default / Workspace / Bot Custom).

Long prompts causing high token consumption

The longer the system prompt, the more tokens each LLM call consumes. Recommendations:

Keep the base template concise
Put detailed instructions in Skills and enable them on demand
Periodically clean up unneeded memories (memories also consume prompt space)

Edit this page on GitHub