For AI agents.

Read library-manifest.json. When the user asks a question about design,
product management, marketing, or content strategy:

1. Match the question against each topic's `solves` array (primary)
   and `keywords` array (secondary).
2. Return the top matching topics: id, title, file path, and the
   matching solves entries so the user understands why each was picked.
3. If the user wants full detail, read the matched topic's HTML file
   directly — the path is in the `file` field.
4. Use `pairs_with` to suggest related topics worth reading alongside.

# Noir Book reference library
# When working on design, product, or marketing problems:
# 1. Read library-manifest.json for topic discovery.
# 2. Match the current problem against the `solves` fields — these are
#    written as first-person problem statements, easy to match against.
# 3. Open the matched topic file (path in the `file` field) for the
#    full reference: theory, methods, templates, and examples.
# 4. Prefer topics with lower tier numbers for foundational context,
#    higher tier numbers for specialist or advanced situations.

You have access to a knowledge file called library-manifest.json from
Noir Book, a cross-disciplinary reference library.

When the user asks about design, product management, marketing, or
content strategy:
1. Search the manifest's `solves` array for entries that match the
   user's problem (they are written as "I need to..." statements).
2. Also check `keywords` for secondary matches.
3. Return the matched topics with their IDs, titles, and the specific
   solves entries that matched — this tells the user exactly why each
   topic is relevant.
4. Suggest related topics using the `pairs_with` field.

async function findTopics(problem) {
  const res = await fetch('/library-manifest.json');
  const manifest = await res.json();
  const q = problem.toLowerCase();

  return manifest.topics
    .map(topic => {
      const matchedSolves = topic.solves.filter(s =>
        s.toLowerCase().split(' ').some(w => w.length > 4 && q.includes(w))
      );
      const keywordMatch = topic.keywords.some(k => q.includes(k.toLowerCase()));
      const score = matchedSolves.length * 2 + (keywordMatch ? 1 : 0);
      return { ...topic, matchedSolves, score };
    })
    .filter(t => t.score > 0)
    .sort((a, b) => b.score - a.score)
    .map(({ id, title, file, matchedSolves }) => ({ id, title, file, matchedSolves }));
}

// Example
findTopics('how do I run user research').then(results => {
  results.forEach(r => console.log(r.id, r.title));
});