✨ feat(graph): make max_words parameter configurable in _synthesize_with_llm

    # Generate answer based on query level and context availability

    if embedding_chunks and query_level in ("advanced", "medium"):

        # Use LLM synthesis with RAG + GraphRAG context

        answer = _synthesize_with_llm(query, embedding_chunks, matching_nodes, edges, max_words=300)

        answer = _synthesize_with_llm(query, embedding_chunks, matching_nodes, edges, max_words=max_words)

    else:

        # Use simple/medium keyword-based answer

        answer = _generate_answer(query, matching_nodes, edges, query_level)

1. Answer the question based ONLY on the provided context

2. If the context doesn't contain enough information, say so

3. Cite specific document IDs and sections when possible

4. Be concise but comprehensive (target: ~{max_words} words)

4. Be concise but comprehensive (target: less than {max_words} words)

Question: {query}

    )

    # Call LLM - calculate roughly 1.3 tokens per word

    int(max_words * 1.3)

    max_tokens = max(4096, int(max_words * 1.3))

    try:

        llm_client = _get_llm_client()

        answer = llm_client.complete(prompt, max_tokens=4096)

        answer = llm_client.complete(prompt, max_tokens=max_tokens)

        return answer

    except Exception as e:

        logger.error(f"LLM synthesis failed: {e}")

# Prompt for concise summary with word count constraint

CONCISE_SUMMARY_PROMPT = """Generate a summary of approximately {target_words} words for this document:

CONCISE_SUMMARY_PROMPT = """Generate a summary of approximately {max_words} words or less for this document:

{content}

def summarize_tdoc(

    document_id: str,

    words: int = 200,

    format: str = "markdown",

    max_words: int = 200,

    fetch_remote: bool = True,

) -> SummarizeResult:

    """Summarize a TDoc by ID.

    Args:

        document_id: Document identifier (e.g., "R1-2400001").

        words: Target word count for summary (default: 200).

        format: Output format - "markdown", "json", or "yaml" (default: "markdown").

        max_words: Maximum word count for summary (default: 200).

        fetch_remote: If True, fetch TDoc metadata from WhatTheSpec if not local.

    Returns:

        ValueError: If format is not supported.

    """

    # Normalize TDoc ID

    normalized_id = document_id.strip().upper()

    # Try to get metadata from WhatTheSpec

    # Generate summary

    truncated_for_summary = _truncate_text(content, SUMMARY_INPUT_LIMIT)

    summary_prompt = CONCISE_SUMMARY_PROMPT.format(

        target_words=words,

        max_words=max_words,

        content=truncated_for_summary,

    )

    try:

        summary = client.complete(summary_prompt, model=config.llm_model, max_tokens=words * 4)

        summary = client.complete(summary_prompt, model=config.llm_model, max_tokens=max_words * 4)

    except Exception as exc:

        msg = f"LLM endpoint is unreachable or misconfigured: {exc}"

        raise LlmConfigError(msg) from exc