Taxonomy Reference

Prompts are classified along five axes. Type says what a prompt IS. Activation captures how it comes alive in an agent's context. The remaining three refine behaviour.

These axes are also the filters on the index. Combine them to pull the slice of the library you want to work with.

A taxonomy of prompts is a map of how people talk to machines. Agentic patterns help them talk to themselves.

Type

What IS this prompt?

System Behavioral rules defining AI identity and persona

Task Immediate work request to complete

Skill Capability with explicit trigger pattern

Reference Documentation, cheatsheets, setup guides

Meta Prompts about prompting conventions

Activation

HOW does it enter context?

Manual Manually placed / Persistent

Invoked Called by name -- slash commands, named tools

Triggered Activates on context match -- file patterns, topics, working state

Reactive Fires on agent lifecycle events -- hooks, guardrails, interceptors

Activity

What WORK does it do? Applies to Task and Skill types.

Task Create

Generate or transform

Task Fix

Correct or validate

Task Understand

Explain or analyze

Constraint

How STRICT are the rules?

Open AI chooses approach freely

Guided Soft preferences given

Bounded Hard rules, some flexibility

Scripted Exact steps prescribed

Scope

How BROAD is the applicability?

Scope Global

All AI interactions

Scope Project

This codebase

Scope Session

This conversation

Scope Atomic

Single use

Saving

Find a prompt worth keeping? The bookmark on its card saves it to /saved, where it sits alongside anything else you've bookmarked.

Landed on a filter set you'll want again? Use Save search next to Clear all in the filter row. The query itself becomes the saved item: your filter set, not a snapshot of today's matches. Find it on /saved beside your saved prompts, and re-run it with a click to see what matches now.