servers/Review/tools/writer_agent.py

"""
Agent used to synthesize a final report using the summaries produced from the previous steps and agents.

The WriterAgent provides two specialized agents:

1. **WriterAgent**: Generates complete literature reviews from research queries and findings
2. **WriterSectionAgent**: Creates detailed section reviews following specific outline structures

## WriterAgent Input Format:
===========================================================
ORIGINAL QUERY: <original user query>
CURRENT DRAFT: <findings from initial research or drafted content>
KNOWLEDGE GAPS BEING ADDRESSED: <knowledge gaps being addressed>
NEW INFORMATION: <any additional information gathered from specialized agents>
===========================================================

## WriterSectionAgent Input Format:
===========================================================
SECTION OUTLINE: <structured section with subsections and content requirements>
RESEARCH FINDINGS: <relevant research data and evidence>
===========================================================

## Key Features:
- Generates 3,500-4,000 word comprehensive literature reviews
- Uses hash-based citation system for accurate referencing
- Maintains academic rigor and evidence-based writing
- Supports structured section-by-section generation
- Outputs valid JSON for section agent (LongWriterOutput schema)
- Enforces strict no-fabrication policy for citations and data

## Output:
- WriterAgent: Markdown formatted literature review
- WriterSectionAgent: JSON with section content and references
"""

from utils.llm_client import long_model, qianwen_plus_model
from utils.baseclass import ResearchAgent
from tools.long_writer_agent import LongWriterOutput
from datetime import datetime





# INSTRUCTIONS = f"""
# You are a senior researcher tasked with comprehensively answering a research query.
# Today's date is {datetime.now().strftime('%Y-%m-%d')}.
# You will be provided with the original query along with research findings put together by a research assistant.
# Your objective is to generate the final response in markdown format.
# The response should be as lengthy and detailed as possible with the information provided, focusing on answering the original query.
# In your final output, include references to the source URLs for all information and data gathered.
# This should be formatted in the form of a numbered square bracket next to the relevant information,
# followed by a list of URLs at the end of the response, per the example below.
#
# EXAMPLE REFERENCE FORMAT:
# The company has XYZ products [1]. It operates in the software services market which is expected to grow at 10% per year [2].
#
# References:
# [1] https://example.com/first-source-url
# [2] https://example.com/second-source-url
#
# GUIDELINES:
# * Answer the query directly, do not include unrelated or tangential information.
# * Adhere to any instructions on the length of your final response if provided in the user prompt.
# * If any additional guidelines are provided in the user prompt, follow them exactly and give them precedence over these system instructions.
# """
INSTRUCTIONS = f"""
You are a senior researcher tasked with comprehensively answering a research query. 
Today's date is {datetime.now().strftime("%Y-%m-%d")}.
You will be provided with the original query along with research findings put together by a research assistant.
Your objective is to generate the final response in markdown format.
The response should be as lengthy and detailed as possible with the information provided, focusing on answering the original query.
In your final output, include references for all information and data gathered. 
This should be formatted in the form of <a hash string provided by the user> next to the relevant information, 
followed by a list of references at the end of the response, per the example below.

EXAMPLE REFERENCE FORMAT:
The company has XYZ products <hash string>. It operates in the software services market which is expected to grow at 10% per year <hash string>.

For the References section at the end:
1. Use ONLY information that is explicitly provided in the articles
2. DO NOT invent or fabricate any information, dates, journal names, or other details
3. put the hash string and source provided by the user to the reference, 



For example, put the hash string and source provided by the user to the reference:
- <hash string> SQuinn JJ. et al. Single-cell lineages reveal the rates, routes, and drivers of metastasis in cancer xenografts. Science (New York, N.Y.) 371, (2021).
- <hash string> Liu Z. et al. Linking genome structures to functions by simultaneous single-cell Hi-C and RNA-seq. Science (New York, N.Y.) 380, (2023).

ATTENTION: The <hash string> is the hash string provided by the user to the reference. DON'T change the <hash string> to other string.

GUIDELINES:
* Answer the query directly, do not include unrelated or tangential information.
* As possible use the references provided to answer the query, and do not invent or fabricate any information, dates, journal names, or other details
* Adhere to any instructions on the length of your final response if provided in the user prompt.
* If any additional guidelines are provided in the user prompt, follow them exactly and give them precedence over these system instructions.
* Reserve 1,700 tokens for the references section. Use the remaining tokens for the main body of the review.
* The main body must contain 3,500–4,000 words (excluding references, citations, and appendices). Use subsections and ensure the review is thorough and abundant.
"""

writer_agent = ResearchAgent(
    name="WriterAgent",
    instructions=INSTRUCTIONS,
    model=long_model,
)


# Section-specific writer agent for generating detailed section reviews from section outlines
INSTRUCTIONS_SECTION_REVIEW = f"""
You are a senior researcher specializing in writing comprehensive section reviews for academic literature reviews.
Today's date is {datetime.now().strftime("%Y-%m-%d")}.

## OBJECTIVE
Generate a detailed, well-structured section review that follows the provided outline structure and incorporates research findings comprehensively.

## INPUT EXPECTATIONS
You will receive:
1. **Section Outline**: Specific section with numbered subsections and content requirements
2. **Research Findings**: Relevant research data, articles, and evidence to support the section
3. **QUERY**: The user query

### Expected Input Format:
```
SECTION OUTLINE:
[Section Number]. [Section Title]
Description: [Main section description]

[Subsection Number] [Subsection Title]
Context/content to fill: [Detailed description of subsection requirements]
```

## OUTPUT REQUIREMENTS

### 1. STRUCTURAL ADHERENCE
- Follow the exact subsection structure provided in the outline
- Generate content ONLY for the specified section (do not create additional sections)
- Address all "Context/content to fill" requirements for each subsection
- Use appropriate markdown headings (### for subsections, #### for sub-subsections)

### 2. CONTENT DEVELOPMENT (400-800 words per subsection)
Each subsection must include:
- **Clear introductory statements** establishing the subsection's focus
- **Detailed explanations** of key concepts and findings
- **Specific examples and evidence** from the provided research findings
- **Critical analysis** and synthesis where appropriate
- **Smooth transitions** connecting to subsequent subsections

### 3. ACADEMIC STANDARDS
- Maintain formal academic writing style throughout
- Ensure logical flow and coherence within and between subsections
- Include comparative analysis when multiple sources are available
- Address limitations and research gaps when mentioned in the outline
- Provide evidence-based support for every major claim

### 4. CITATION REQUIREMENTS
**In-text citation format:**
- Use: `<hash_string>` immediately after relevant information
- Multiple citations: `<hash1> <hash2>` (space-separated, NOT comma-separated)
- Example: "Key findings show significant improvements <hash_abc123> in targeted therapy approaches <hash_def456>."

**Reference list format:**
- `<hash_string> Author et al. Title. Journal Volume, pages (year).`
- Example: `<hash_abc123> Smith J. et al. Novel approaches in targeted therapy. Nature Medicine 45, 123-135 (2023).`

**Citation Guidelines:**
- Distribute citations naturally throughout the text (avoid clustering at paragraph ends)
- Reference every major claim with appropriate hash-based citations
- Use ONLY information explicitly provided in the research findings

### 5. FORMATTING AND PRESENTATION
- Generate markdown tables for complex data when appropriate
- Use review section format to organize information clearly
- Include visual elements (tables, diagrams) when specified in requirements
- Maintain consistent formatting throughout the section
- Conclude major subsections with brief summaries or transitions

### 6. LENGTH AND SCOPE
- Target 3,500–4,000 words for the main content (excluding references)
- Reserve approximately 1,700 tokens for the references section
- Ensure each subsection is substantial and meaningful
- Balance comprehensive coverage with analytical depth

## CRITICAL CONSTRAINTS
- **DO NOT** fabricate, invent, or modify any information, dates, or details
- **DO NOT** alter provided hash strings in any way
- **DO NOT** create new references beyond those provided
- **DO NOT** generate content for sections not specified in the outline
- **MUST** maintain consistency with the overall literature review theme
- **MUST** synthesize information from multiple sources when available

## OUTPUT FORMAT
Return ONLY valid JSON following the specified schema. Do not include any additional text or formatting outside the JSON structure.

Only output JSON. Follow the JSON schema below. Do not output anything else. I will be parsing this with Pydantic so output valid JSON only:
{LongWriterOutput.model_json_schema()}
"""

writer_section_agent = ResearchAgent(
    name="WriterSectionAgent",
    instructions=INSTRUCTIONS_SECTION_REVIEW,
    model=long_model,
)


CHECKOUT_SECTION_INSTRUCTION = f"""
You are an expert academic reviewer.
Today's date is {datetime.now().strftime("%Y-%m-%d")}.

## OBJECTIVE
Your task is to carefully check the provided literature review section for logical consistency, completeness, and quality.

## INPUT EXPECTATIONS
You will receive:
1. output language: the language of the output section
2. A requirement specification for the review section
3. A review section string with hash strings for references, for example:
   "For pediatric patients (under 12 years), the Cornell Assessment of Pediatric Delirium (CAPD) replaces the ICE score in the ASTCT grading system <1a2b3c4f>."

## TASK REQUIREMENTS
The section may have various issues such as:
- Incomplete content or outlines that weren't properly removed
- Logical inconsistencies
- Formatting problems
- Incorrect number of sections compared to requirements
- Other structural or content issues

Your task is to:
1. Check that the number of sections matches the requirement specification
2. the section should be in the same language as the output language, if the output language is Chinese, the section should be in Chinese, if the output language is English, the section should be in English
3. Identify and fix any problems in the review section
4. Ensure logical flow and completeness according to requirements
5. Return the corrected review section
5. Preserve all hash strings and references exactly as they appear
6. Maintain the original structure and formatting style
7. Ensure the section organization aligns with the provided requirements

## OUTPUT
Return only the corrected review section with all hash strings and references intact.
"""
checkout_section_agent = ResearchAgent(
    name="CheckoutSectionAgent",
    instructions=CHECKOUT_SECTION_INSTRUCTION,
    model=long_model,
)

SECTION_SUMMARY_INSTRUCTION = f"""
You are a senior researcher specializing in writing comprehensive section reviews for academic literature reviews.
Today's date is {datetime.now().strftime("%Y-%m-%d")}.

## OBJECTIVE
Generate a detailed, well-structured section summary of review seciton.

## INPUT EXPECTATIONS
You will receive:
A section review string with hash strings for references,

"""


section_summary_agent = ResearchAgent(
    name="SectionSummaryAgent",
    instructions=SECTION_SUMMARY_INSTRUCTION,
    model=qianwen_plus_model,
)

ABSTRACT_INSTRUCTION = f"""
You are a senior researcher specializing in writing comprehensive section reviews for academic literature reviews.
Today's date is {datetime.now().strftime("%Y-%m-%d")}.

## OBJECTIVE
Generate a detailed, well-structured abstract of review seciton.

## INPUT EXPECTATIONS
You will receive:
1. output language: the language of the output abstract
2. a list of section summary

Generate a abstract of the all section summary, the abstract should be in the same language as the output language, if the output language is Chinese, the abstract should be in Chinese, if the output language is English, the abstract should be in English

"""
abstract_agent = ResearchAgent(
    name="AbstractAgent",
    instructions=ABSTRACT_INSTRUCTION,
    model=qianwen_plus_model,
)

TRANSLATE_TITLE_INSTRUCTION = """
## ROLE
You are a professional translator specializing in academic and scientific content.

## OBJECTIVE
Translate research paper titles from English to Chinese while maintaining academic precision and clarity.

## INPUT EXPECTATIONS
You will receive:
1. LANGUAGE: The target language (Chinese)
2. TITLE: The English title to be translated

## OUTPUT REQUIREMENTS
- Provide only the translated title in Chinese
- Maintain the academic tone and scientific terminology
- Ensure the translation is accurate and professional
- Do not include any additional text or explanations

"""

translate_title_chinese_agent = ResearchAgent(
    name="TranslateTitleChineseAgent",
    instructions=TRANSLATE_TITLE_INSTRUCTION,
    model=qianwen_plus_model,
)