Most "prompt frameworks" are branding. A few patterns survive contact with real work. The one Illumora teaches first is four blocks: role, task, constraints, output.
The four blocks
Role
Who should the model act as — and for whom? Be specific enough to set taste and expertise, not theatrical.
Task
The single job. One primary verb. If you have three jobs, write three prompts or sequence them deliberately.
Constraints
Hard edges: length, audience, forbidden claims, sources, compliance. Constraints are where quality lives.
Output
The shape you will paste, ship, or parse. Sections, JSON keys, bullet counts, or "reply only with the email body."
Why it travels
- ChatGPT, Claude, and Gemini all reward explicit structure. Vendor quirks (XML tags for Claude, schemas for OpenAI) sit *inside
- this scaffold — they do not replace it.
Technique deep-dives
The registry turns this into named techniques: Role + objective, Hard constraints, Output schema, and vendor-specific variants. Browse the technique library when you want the research trail behind each move.
Related techniques
- Hard constraintsExplicit boundaries the model must not violate.
- Output schemaForce a reusable structure the user can copy or parse.
- Role + objectiveState who the model is and what success looks like before the task.
- PTCF framingPersona → Task → Context → Format — Gemini-aligned structure.