"""Structured LLM prompts for document enrichment.

Provides the system prompt that instructs the LLM to return structured JSON
with entities, categories, tags, and action items.
"""

ENRICHMENT_SYSTEM_PROMPT = """\
You are a knowledge management assistant. Analyze the following text extracted \
from a document and return structured metadata as JSON.

Return a JSON object with exactly these fields:
- "title": A concise, descriptive title for this note (max 80 chars)
- "category": One of "meeting", "project", "decision", "inbox"
- "tags": List of relevant topic tags (lowercase, no # prefix)
- "entities": List of detected entities, each with:
  - "type": One of "person", "project", "date", "action_item"
  - "value": The extracted text value
  - "confidence": Float between 0.0 and 1.0 indicating detection confidence
  - "position": Approximate character offset where entity appears (integer or null)
- "action_items": List of action items, each with:
  - "description": What needs to be done
  - "assignee": Person responsible (string or null)
  - "deadline": Date in ISO 8601 format, e.g. "2024-03-22" (string or null)
- "summary": A 1-2 sentence summary of the document (string or null)

Rules:
- Be precise with entity detection. Only mark entities you are confident about.
- For people, use their full name as the value.
- For projects, use the project name as commonly referenced.
- For dates, use ISO 8601 format (YYYY-MM-DD) as the value.
- Assign confidence scores honestly: 0.9+ for clearly stated entities, 0.7-0.9 for \
likely entities, below 0.7 for uncertain ones.
- Category "meeting" for meeting notes, agendas, minutes.
- Category "project" for project-specific documentation.
- Category "decision" for recorded decisions, ADRs.
- Category "inbox" for anything that doesn't clearly fit the above.

Context: This is a personal knowledge base for a project manager working across \
multiple business projects.{context_section}
"""

STRICT_RETRY_PROMPT = """\
Your previous response was not valid JSON. Please respond with ONLY a valid JSON \
object matching the schema described in the system prompt. No markdown, no code fences, \
no explanation — just the raw JSON object.
"""


def build_system_prompt(
    known_projects: list[str] | None = None,
    known_people: list[str] | None = None,
) -> str:
    """Build the system prompt with optional context about known entities.

    Args:
        known_projects: List of known project names for context.
        known_people: List of known people names for context.

    Returns:
        Complete system prompt string.
    """
    context_parts: list[str] = []

    if known_projects:
        projects_str = ", ".join(known_projects)
        context_parts.append(f"Known projects: {projects_str}.")

    if known_people:
        people_str = ", ".join(known_people)
        context_parts.append(f"Known people: {people_str}.")

    if context_parts:
        context_section = "\n" + " ".join(context_parts)
    else:
        context_section = ""

    return ENRICHMENT_SYSTEM_PROMPT.format(context_section=context_section)


def build_user_prompt(extracted_text: str) -> str:
    """Build the user message containing the extracted document text.

    Args:
        extracted_text: The raw text extracted from the document.

    Returns:
        User prompt string.
    """
    return f"Analyze this document and return the structured JSON:\n\n{extracted_text}"