Prompting visual guide

An effective prompt contract

The layers that turn a vague request into an auditable task contract across providers.

Editable sourceMermaidOpen full-size SVG ↗Download .mmd source

A vertical stack shows Authority at the top, followed by Objective, Trusted context, Untrusted input boundary, Constraints and tools, Output contract, Success criteria, and Verification and escalation at the bottom. A side label marks untrusted input as data rather than instructions. — A provider-neutral prompt contract separates authority and trusted context from untrusted input, then defines output and verification.

Read the visual

What this shows

A strong prompt distinguishes authority, objective, trusted context, untrusted input, constraints, output shape, success criteria, and verification. This structure works across providers because it describes the work rather than relying on model-specific phrasing.

Carry this forward

Key takeaways

Treat retrieved or user-supplied content as data unless it has explicit authority.
Specify the output contract and success criteria before asking for generation.
Include verification and escalation rules for work that can fail or cause harm.