Keep the goal. Adapt the instruction.
Use the same business goal with instructions suited to the selected tool and available context.
Inspect the destination’s inputs and tools
A model-aware prompt accounts for the destination’s supported inputs, response format and task environment. A text chat, image generator and coding assistant cannot all execute the same instructions in the same way.
| Check | Why it matters |
|---|---|
| Available input | A file named in the prompt may not be attached or accessible |
| Output format | A requested schema or artifact may need tool support |
| Tool access | Browsing or execution must actually be available |
| Review criteria | The result must be checked against the task, not the model name |
A portable task with explicit dependencies
Make the unavailable-tool behavior part of the instruction instead of letting the model pretend.
Compare the supplied product specifications using the same criteria for each. If browsing is available and authorized, verify current prices from official pages and include dates. Otherwise, leave current price unverified and compare only the supplied specifications. Return a compact table and explain which missing facts would change the recommendation.
Compatibility is not identical output
Different models may produce different answers from the same prompt. A supported prompt target is not necessarily a native connector, and it does not include the destination’s subscription. Check the current model list and connection method.
Common questions
Does one prompt work everywhere?
The task can often transfer. Input formats, tool access, context limits and behavior still differ, so review the instruction and result on the chosen destination.
Sources and review
Reviewed