PM-301g · Module 1

Taxonomy Design

4 min read

Taxonomy is how you find things. A prompt library with excellent metadata but poor taxonomy is a database that requires you to know the exact prompt ID to retrieve anything. Two taxonomy patterns dominate: hierarchical and flat-tag. Both are valid. The choice depends on the organization's retrieval patterns and the library's expected size.

Hierarchical taxonomy organizes prompts in a tree: domain → task → subtask. For example: sales → email → cold-outreach, or operations → summarization → meeting-notes. Hierarchical taxonomy scales well for large libraries because it limits the scope of each search. Looking for a meeting summary prompt? Navigate to operations → summarization → and you are looking at 12 prompts, not 300. The cost is maintenance: when a prompt spans two domains or a new domain emerges, the hierarchy must be updated.

Flat tag systems are more flexible. A prompt can carry multiple tags across multiple dimensions: ["sales", "email", "cold-outreach", "short-form", "b2b"]. Any tag combination retrieves matching prompts. Flat tags work well for smaller libraries and teams with diverse, overlapping use cases. The cost is inconsistency: without tag governance, the same concept accumulates multiple spellings (email, emails, e-mail) and searches miss results.

Do This

  • Use hierarchical taxonomy for libraries over 100 prompts
  • Use flat tags for libraries under 50 prompts or with highly cross-domain use cases
  • Publish a canonical tag vocabulary and enforce it — no freeform tags
  • Support both: hierarchy for primary classification, flat tags for secondary attributes

Avoid This

  • Let contributors invent new tags without governance — tag proliferation destroys recall
  • Use freeform search as a substitute for taxonomy — it does not scale
  • Flatten a large library — navigating 300 prompts by tag search alone is slow
  • Over-engineer taxonomy for a library of 20 prompts — the cost outweighs the benefit