Frameworks
A framework is a checklist of the things a model cannot guess. Ten of them below, from three-slot starters to six-slot workhorses — with the two that a vendor actually publishes marked as such.
RTF · Role · Task · Format
The three-line starter. Say who the model is, what it must do, and what shape the answer takes.
TAG · Task · Action · Goal
The smallest useful structure. Three sentences that separate the subject, the operation, and the reason.
APE · Action · Purpose · Expectation
TAG with the emphasis moved to the finish line: what to do, why, and how you will judge the result.
RACE · Role · Action · Context · Expectation
RTF plus the two things RTF is missing: background, and a definition of done.
CRAFT · Context · Role · Action · Format · Tone
The general-purpose workhorse. Five slots that between them cover most of what a model needs to stop guessing.
GCSE · Goal · Context · Source · Expectations
Microsoft’s own framework for Copilot. The only common framework with a first-class slot for which of your files to read.
TCREI · Task · Context · References · Evaluate · Iterate
Google’s framework, and the only common one that treats prompting as a loop rather than a single message.
RISEN · Role · Input · Scenario · Expectation · Nuance
The heavyweight. Built for multi-step work where the edge cases matter as much as the main path.
CO-STAR · Context · Objective · Style · Tone · Audience · Response
Six slots, three of them about voice. The framework to reach for when who is reading this matters as much as what it says.
CLEAR · Concise · Logical · Explicit · Adaptive · Reflective
Not a template — a rubric. Five principles for judging a prompt you have already written. This site’s quality meter scores against them.
PICSE · Persona · Intention · Criteria · Sources · Examples
The UK Government’s framework, and the only one here with both source material and worked examples as required slots.
TELeR · Turn · Expression · Level of detail · Role
Not a template but a measuring scale: a peer-reviewed answer to "how detailed does a prompt need to be", with seven rungs instead of an adjective.
ReAct · Thought · Action · Observation
Not a message but a loop. The model reasons, calls a tool, reads what came back, and re-plans — the ancestor of every agent framework shipping today.
CREATE · Character · Request · Examples · Adjustments · Type of output · Extras
The most iteration-aware of the templates: it has a dedicated slot for what the model must stop doing, and another for instructions about the conversation itself.
PREPARED · Propose · Role · Explicit · Presentation · Ask · Rate · Emotions · Diversity
Eight levers rather than eight steps, and the only common framework with a bias check and a self-review built in.