DORA Template Configuration

The DORA template configuration file (.json) defines the structure, behavior, and content generation logic of a document template. It is essential for guiding the DORA kernel through document creation, including user inputs, section structure, interdependencies, and tool integrations.

💡 Configuration Highlights

• Modularity – Each section is independently controlled with specific prompts and logic.
• Tool Integration – Tools such as literature search and omics analysis enhance scientific quality.
• Consistency – Shared memory enforces narrative coherence across sections.
• Instruction Precision – Prompts ensure academic tone, structure, and non-redundant output.

📚 Glossary

Term	Definition
slug	A unique, machine-readable identifier for each section or input
prompt	Instructional text given to the AI model to guide content generation
shared_memory	Mechanism that allows one section to recall content from others implicitly
depends_on	Explicit dependency on other section outputs via placeholder references
tool	External AI agent (e.g., PubMed search) used for data augmentation
plan_prompt	Prompt to create an outline/structure before full text generation
placeholder	A marker within prompts replaced with user input, tool data, or other content at runtime

🔑 Mandatory Top-Level Keys

Field	Description	Data Type
template_type	High-level category for grouping templates in the UI. Must match predefined values such as Original research, Business report, Literature review, etc.	string
template_description	Brief explanation of the purpose and content of the template.	string
sections	Defines the structure of the document using unique slugs as keys. Each value configures a section’s prompt, tools, and rules.	object (dictionary)
user_inputs	Input fields the user must complete before document generation. Referenced in section prompts via [placeholder].	array of objects
shared_memory	Describes memory flow between sections to ensure consistent context.	object (dictionary)

📝 User Input Object

These define the fields users complete prior to generation. They appear in the DORA interface under Document Settings.

Key	Type	Description
slug	string	Machine-readable ID for the field (e.g., topic)
type	string	Input type. Currently supports "free_text_input"
text_limit	integer	Maximum allowed characters
display_name	string	User-facing label for the input field
display_size	string	Field size in UI (e.g., "large", "small")
default_value	string	Pre-filled value for user guidance

Screenshot to be added

🧩 Section Object

Each section of the document is represented as an object in the sections dictionary.

Key	Type	Description
slug	string	Unique ID (e.g., main_text)
title	string	Title shown to the user
prompt	string	Main instruction for text generation (uses placeholders)
expected_output_instructions	string	Format and tone guidance for the output
is_title	boolean	If true, marks this as the document title
tools	array	List of tool slugs available for this section
depends_on	array	List of section slugs this section depends on
plan_prompt	string	Prompt to create an outline/plan (if needed)
requires_plan	boolean	If true, section generation starts with a plan
dynamic_template_subsection	boolean	Whether the section supports dynamic subsections
display_on_plan_overview	boolean	Controls whether the section is shown in the research plan UI
display_name_on_plan_overview	string or null	Optional custom display name in plan view
custom_parts	(empty)	Reserved for injected interface values
sub_sections	array (unused)	Reserved for future use

🧱 Sections Overview

Below are examples of common section configurations. You can explore the configuration example on the admin page.

1. Title

Property	Value
slug	title
prompt	Generates a single-line title
expected_output_instructions	"One single-line title."
is_title	true
tools	None

2. Abstract

Property	Value
slug	abstract
prompt	Summarizes key points and significance
expected_output_instructions	150–250 word paragraph, no references
tools	None

3. Introduction

Property	Value
tools	pubmed_abstract_similarity_search_tool, web_search_tool
prompt	Explains the topic, narrowing focus and setting context
expected_output_instructions	300–400 words; no repetitive phrasing

4. Main Text

Property	Value
requires_plan	true
dynamic_template_subsection	true
tools	All scientific tools enabled
plan_prompt	Outputs a 2-3 section structured outline
prompt	Executes the plan with depth and clarity
expected_output_instructions	~7000 words; one cohesive narrative, no duplication

5. Conclusion

Property	Value
tools	Literature tools
prompt	Summarizes insights and future directions
expected_output_instructions	150–200 words; single paragraph, no repeated ideas

🧠 Shared Memory and Dependencies

1. Shared Memory

Defined in the shared_memory object:

{
    "title": [
        "main_text"
    ],
    "abstract": [
        "main_text"
    ],
    "main_text": [],
    "conclusion": [
        "main_text"
    ],
    "introduction": [
        "conclusion"
    ]
}

• Enables automatic context transfer between sections.
• Improves narrative flow and avoids content duplication.
• Memory is unidirectional.

2. Depends On

Used for deterministic content reuse from another section. Example:

"depends_on": ["Network analysis"]

Prompt must include a placeholder like:

[Network analysis.results]

This will inject the entire generated output from that section.

3. Comparison: Shared Memory vs Depends On

Aspect	Shared Memory	Depends On
Purpose	Recall context	Insert section output verbatim
Location in Config	Top-level key	Inside section object
Placeholder Required	❌ No	✅ Yes ([SectionSlug.results])
Best For	Summaries, intros, conclusions	Analytical reuse, citations, methods
Output Control	Implicit	Explicit
Prevents Duplication	✅ Yes	❌ Only if placeholders are used correctly

🔧 Tool Selection and Prompt Design

DORA dynamically integrates scientific tools into the generation process.

Prompt Placeholders

Each prompt that supports tools should include:

• {activated_tools|<group>|names} – Inserted list of active tool names
• {activated_tools|<group>|instructions} – Combined instructions from active tools

Example

Prompt before substitution:

Use {activated_tools|scientific_search_tools|names} when needed.

{activated_tools|scientific_search_tools|instructions}

After substitution (tools enabled):

Use pubmed_abstract_similarity_search_tool and web_search_tool when needed.

Instructions for pubmed_abstract_similarity_search_tool: ...

Instructions for web_search_tool: ...

Admin Tool Configuration

Tool metadata is stored in:

/admin/kernel/toolpromptinstructiongroup/

Example:

{
    "name": "scientific_search_tools",
    "tools": [
        {
            "name": "pubmed_abstract_similarity_search_tool",
            "instruction": "..."
        },
        {
            "name": "web_search_tool",
            "instruction": "..."
        }
    ]
}

Validation Rules

• {...|names} and {...|instructions} must appear together
• Tools listed must exist and include valid instructions
• At least one tool from the group must be selected if placeholders are used

📥 How to Load a Custom Template to Admin Page

[List of screenshots from OS version]

Validation Rules

1. Slugs must be unique across user inputs and sections.
2. Only one section can be displayed on the plan overview.
3. Only one section can be marked as a title.
4. All dependencies (depends_on) must reference existing slugs.
5. Shared memory and dependency maps are merged and checked for conflicts.
6. Tool placeholders must be paired and backed by actual tools in the section.
7. Prompt placeholders must:
8. Match valid user inputs or section dependencies.
9. Avoid referencing keys not declared in depends_on.

Assign Tools and Instructions [todo]

• Navigate to Tool Prompt Instruction Group
• Define instruction templates for AI tool integration
• These will be inserted dynamically into prompts during generation