app/core/srs_assembler.py

"""
SRS Assembler

Deterministic assembly of SRS document from agent outputs, followed by
a single LLM cross-validation pass. Replaces the spec_coordinator agent.
"""

import logging
from datetime import datetime

from langchain_core.prompts import ChatPromptTemplate

from .llm_factory import get_chat_model
from .schemas import TeamRole

logger = logging.getLogger("srs_assembler")

# Static documentation guidelines appendix (replaces technical_writer agent)
STATIC_DOC_APPENDIX = """# Documentation Guidelines

## Recommended Documentation Structure

### For Development Teams
1. **README.md** - Project overview, setup instructions, quick start
2. **ARCHITECTURE.md** - System architecture, component diagrams, technology decisions
3. **API.md** - API documentation, endpoint reference, authentication
4. **DEPLOYMENT.md** - Deployment guide, environment configuration, CI/CD pipeline
5. **CONTRIBUTING.md** - Coding standards, PR process, code review guidelines

### For Product Teams
1. **PRD.md** - Product requirements, user stories, acceptance criteria
2. **ROADMAP.md** - Feature roadmap, priorities, timeline
3. **DECISIONS.md** - Architecture decision records (ADRs)

### Documentation Standards
- Use Markdown format for all documentation
- Keep documentation close to code (in repository)
- Update documentation with every code change
- Use diagrams (Mermaid, PlantUML) for complex concepts
- Maintain a single source of truth

## Diátaxis Framework
Organize documentation into four categories:
1. **Tutorials** - Learning-oriented, step-by-step guides
2. **How-to Guides** - Problem-oriented, practical instructions
3. **Reference** - Information-oriented, technical specifications
4. **Explanation** - Understanding-oriented, conceptual background
"""


def assemble_srs(
    all_outputs: dict[str, str],
    knowledge_graph: dict | None = None,
    contradictions: list[dict] | None = None,
) -> str:
    """
    Deterministically assemble SRS document from agent outputs.

    Args:
        all_outputs: Dictionary mapping role names to their markdown output
        knowledge_graph: Optional merged knowledge graph dict for entity context
        contradictions: Optional list of cross-agent contradiction dicts

    Returns:
        Complete SRS document as markdown string
    """
    sections = [
        ("1. Product Requirements", all_outputs.get("product_owner", "")),
        ("2. Functional Requirements", all_outputs.get("business_analyst", "")),
        ("3. Technical Architecture", all_outputs.get("solution_architect", "")),
        ("4. Data Architecture", all_outputs.get("data_architect", "")),
        ("5. Security Requirements", all_outputs.get("security_analyst", "")),
        ("6. User Experience Design", all_outputs.get("ux_designer", "")),
        ("7. API Specifications", all_outputs.get("api_designer", "")),
        ("8. Testing Strategy", all_outputs.get("qa_strategist", "")),
        ("9. DevOps & Infrastructure", all_outputs.get("devops_architect", "")),
    ]

    # Build table of contents
    toc_lines = ["# Software Requirements Specification\n"]
    toc_lines.append(
        f"**Generated:** {datetime.utcnow().strftime('%Y-%m-%d %H:%M:%S UTC')}\n"
    )
    toc_lines.append("---\n")
    toc_lines.append("## Table of Contents\n")
    for title, _ in sections:
        anchor = title.lower().replace(" ", "-").replace(".", "")
        toc_lines.append(f"- [{title}](#{anchor})")
    toc_lines.append("- [A. Documentation Guidelines](#a-documentation-guidelines)")
    if contradictions:
        toc_lines.append("- [B. Consistency Warnings](#b-consistency-warnings)")
    toc_lines.append("\n---\n")

    # Build document body
    body_parts = ["\n".join(toc_lines)]

    for title, content in sections:
        if content and content.strip():
            body_parts.append(f"\n## {title}\n\n{content}\n")

    # Add consistency warnings section if contradictions exist
    if contradictions:
        from .consistency_validator import ConsistencyValidator

        validator = ConsistencyValidator()
        warnings = validator.format_warnings_section(contradictions)
        if warnings:
            body_parts.append(f"\n{warnings}\n")

    # Add static documentation appendix
    body_parts.append(f"\n## A. Documentation Guidelines\n\n{STATIC_DOC_APPENDIX}\n")

    return "\n".join(body_parts)


async def validate_srs(srs_document: str, all_outputs: dict[str, str]) -> str:
    """
    Cross-validate assembled SRS for internal consistency.

    Single LLM pass to check for contradictions, gaps, and completeness.

    Args:
        srs_document: The assembled SRS document
        all_outputs: Original agent outputs for reference

    Returns:
        Validation report string
    """
    validator_prompt = """You are a senior technical reviewer validating a Software Requirements Specification.

Review the assembled SRS below for:

1. **Internal Consistency**: Are there contradictions between sections? (e.g., architecture says PostgreSQL but data model says MongoDB)
2. **Completeness**: Are all required sections present and substantive?
3. **Traceability**: Do requirements link to architecture, tests, and API specs?
4. **Gaps**: Is there missing critical information?

## SRS Document

{srs_document}

## Your Task

Return a validation report in this format:

```markdown
## Validation Report

### Consistency Check
[PASS/FAIL] - [Details]

### Completeness Check
[PASS/FAIL] - [Details]

### Traceability Check
[PASS/FAIL] - [Details]

### Identified Gaps
- [Gap 1]
- [Gap 2]

### Overall Assessment
[VALID / NEEDS_REVISION]
```

Be specific and actionable. If the SRS passes all checks, return "VALID" as the overall assessment."""

    llm = get_chat_model(
        role=TeamRole.SPEC_COORDINATOR, temperature=0.2, max_tokens=2048
    )

    prompt_template = ChatPromptTemplate.from_messages(
        [
            ("system", validator_prompt),
            ("human", "Please validate the SRS document above."),
        ]
    )

    chain = prompt_template | llm

    try:
        response = await chain.ainvoke({"srs_document": srs_document[:15000]})
        return response.content if hasattr(response, "content") else str(response)
    except Exception as e:
        logger.error(f"SRS validation failed: {e}")
        return f"## Validation Report\n\n**Status:** Validation failed due to error: {e}\n\n**Recommendation:** Review SRS manually."