core/linter.py

"""Wiki linter — health-checks the knowledge base for issues."""

import anthropic
import json

LINT_SYSTEM_PROMPT = """You are auditing a nursing knowledge wiki for the Nursing Citizen Development Organisation.
Perform a thorough health check and identify:

1. **Contradictions**: Where articles disagree with each other or with current evidence
2. **Stale content**: Claims that may be outdated (check against NMC 2018, NICE, NHS guidelines)
3. **Orphan articles**: Articles with few or no backlinks (disconnected from the rest of the wiki)
4. **Missing cross-references**: Where articles should link to each other but don't
5. **Clinical safety gaps**: Important safety information that is absent or incomplete
6. **Missing articles**: Important nursing topics not yet covered (suggest 3-5 new articles)
7. **Evidence gaps**: Claims without adequate citation

Return a structured JSON report:
{
  "overall_health": "Good/Fair/Needs attention",
  "total_issues": number,
  "issues": [
    {
      "type": "contradiction|stale|orphan|missing_link|safety_gap|evidence_gap",
      "severity": "high|medium|low",
      "article": "article_slug or 'wiki-wide'",
      "description": "Clear description of the issue",
      "recommendation": "Specific action to fix it"
    }
  ],
  "suggested_new_articles": [
    {
      "title": "Suggested Article Title",
      "category": "category",
      "rationale": "Why this article is needed",
      "key_topics": ["topic1", "topic2"]
    }
  ],
  "strengths": ["What the wiki does well"],
  "summary": "Brief overall assessment"
}
"""


def lint_wiki(client: anthropic.Anthropic, articles: dict, index_summary: str,
              model: str = "claude-sonnet-4-6") -> dict:
    """
    Run a health check on the wiki.

    Returns a structured report with issues and recommendations.
    """
    # Build article summaries for the linter
    article_summaries = []
    for slug, art in articles.items():
        # Extract backlinks from content
        import re
        backlinks = re.findall(r'\[\[([^\]]+)\]\]', art["content"])
        article_summaries.append({
            "slug": slug,
            "title": art["title"],
            "category": art["category"],
            "tags": art.get("tags", []),
            "last_updated": art.get("last_updated", "unknown"),
            "word_count": len(art["content"].split()),
            "backlinks": backlinks,
            "sources": art.get("sources", []),
            "content_preview": art["content"][:600],
        })

    prompt = f"""## Wiki Index
{index_summary}

## Article Summaries
{json.dumps(article_summaries, indent=2)[:12000]}

Please perform a thorough health check of this nursing knowledge wiki.
Return valid JSON only, no markdown fences."""

    response = client.messages.create(
        model=model,
        max_tokens=3000,
        system=LINT_SYSTEM_PROMPT,
        messages=[{"role": "user", "content": prompt}],
    )

    raw = response.content[0].text.strip()
    if raw.startswith("```"):
        raw = raw.split("\n", 1)[1]
        if raw.endswith("```"):
            raw = raw.rsplit("```", 1)[0]

    return json.loads(raw)


def generate_missing_article(client: anthropic.Anthropic, title: str, category: str,
                              key_topics: list, existing_index: str,
                              model: str = "claude-sonnet-4-6") -> dict:
    """Generate a new article for a topic identified as missing by the linter."""
    import datetime

    prompt = f"""Write a comprehensive nursing knowledge wiki article on: **{title}**

Category: {category}
Key topics to cover: {', '.join(key_topics)}

Existing wiki context (do not duplicate, but do cross-reference):
{existing_index}

The article should:
- Be clinically accurate and evidence-based (NMC 2018, NICE, NHS, BNF standards)
- Use UK spellings and references
- Include backlinks to related articles using [[Article Title]] format
- Have clear sections with ## headers
- Include a References section
- Be suitable for student nurses

Return the full markdown article content only (no JSON, just the markdown)."""

    response = client.messages.create(
        model=model,
        max_tokens=3000,
        messages=[{"role": "user", "content": prompt}],
    )

    content = response.content[0].text.strip()
    slug = title.lower().replace(" ", "_").replace("-", "_")
    slug = "".join(c for c in slug if c.isalnum() or c == "_")

    return {
        "slug": slug,
        "title": title,
        "category": category,
        "tags": key_topics[:5],
        "content": content,
        "last_updated": datetime.date.today().isoformat(),
        "sources": ["AI-generated from lint suggestion"],
    }