Write an agent prompt with voice and keyboard together

Vibe Pad combines programmable physical controls with a separate microphone for a chosen dictation application. Voice and keyboard input can share the same prompt draft. Speaking is useful for describing intent; typing or copying is often clearer for exact filenames, symbols and commands. Assign each part of the prompt to the input method that makes it easiest to verify.

What should you say aloud first?

Capture the desired outcome and the relevant context in ordinary language. Explain what prompted the task and what a successful result would look like. If you are considering alternatives, say which choice is settled and which still needs investigation.

Keep the transcript in an editable field. Spoken self-corrections can leave both the old and new instructions in place, so do not treat the first transcription as ready to send. End capture deliberately before reorganising the text unless your software explicitly supports editing while listening.

Which details are better inserted exactly?

Add file paths, identifiers, URLs and code fragments from the relevant source. Check that the inserted reference belongs to the current project and version. A perfectly copied path from the wrong workspace still gives the agent misleading context.

Use the keyboard to structure the draft when speaking formatting commands feels slower than editing. Turn a long paragraph into a few clear requirements, and remove temporary notes that no longer belong. The purpose is a coherent instruction, not a demonstration that every character arrived through the microphone.

How do you avoid losing the insertion point?

After visiting another window to copy a reference, return to the intended field and inspect the cursor before resuming speech. Some transcription tools deliver text to the foreground application, while others retain a draft panel. Test the behaviour of your actual software with neutral content.

Keep recording-stop and prompt-submit actions separate in the keypad mapping. A final result may arrive after capture ends, and you need a chance to see where it landed. Do not attach an automatic send action merely to save one deliberate review step.

What should the combined review cover?

Read the entire prompt after joining the spoken explanation and exact references. Check negations, scope boundaries and any assumptions presented as facts. Ensure the technical fragments still make sense inside the edited prose rather than appearing as disconnected tokens.

State a concrete way to assess the result, then submit from the visible destination when the draft is ready. The Vibe Pad configuration details describe the input components; a useful hybrid workflow comes from choosing where each method reduces effort without removing review.