> ## Documentation Index
> Fetch the complete documentation index at: https://docs.voqo.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Refine and Test Prompts

> Use prompt tools to improve agent behavior before deploying prompt changes.

## Audience

* Operators iterating on call behavior and tone
* Admins maintaining quality across multiple agents

## Prerequisites

* Access to prompt tools in workspace
* Existing agent prompt draft
* Test scenario for validation

## Plan and role requirements

* Prompt tooling availability may depend on plan.
* Editing production prompts should be limited to trusted roles.

## Steps

### 1) Draft or import prompt

1. Open prompt tools.
2. Paste or draft your prompt content.
3. Define desired behavior outcomes.

<Note>
  **Already have a custom prompt?** When you sync an agent whose prompt wasn't built in Prompt Studio, it's **rewritten into Voqo's standard format** — only your call-handling flows carry over; bespoke wording, tone, and special instructions are not retained. Prompt Studio shows a warning when this applies. Your previous prompt is kept in version history (you can roll back), and you can review the full result in the **Advanced** tab before you save.
</Note>

### 2) Refine with prompt assistant

1. Run refinement suggestions.
2. Apply improvements for clarity and constraints.
3. Save refined output for test.

### 3) Test in writing

1. Open the **Test** tab. Your agent opens the conversation with its greeting first — just like a real call, where it speaks before the caller does. Type a caller message to continue.
2. Compare output to expected behaviour.
3. Iterate until response quality is stable.

The test applies the same **date and time context** and **response-style rules** your agent receives on a live call, so its answers about dates, scheduling, and tone match production. The greeting is shown as the agent's first turn for realism, but — exactly as on a real call — it is spoken at the start and is not part of the prompt the agent reasons over.

If your agent uses a **personalised greeting**, the test shows your base opening line as a representative example; the live greeting is tailored to each caller. Caller details and callback context only apply to real return calls and are not simulated here.

### 4) Deploy to agent

1. Update the target agent prompt.
2. Run controlled call test.
3. Verify no regression in key workflows.

## When to use prompts vs skills

* Use **prompts** for language, behavior, and instruction quality.
* Use **skills** for reusable structured capabilities and tool-like functions.

## Troubleshooting

### Prompt quality regresses after update

* Roll back to known stable prompt version.
* Re-test with same scenario set.
* Apply smaller incremental changes.

### Tool output differs from call behavior

* Confirm deployed prompt matches tested prompt.
* Check skills/functions attached to agent that may alter behavior.

If unresolved, contact [support@voqo.ai](mailto:support@voqo.ai) with workspace ID, agent ID, prompt version context, and failing scenario.

## Related docs

* [Create and Bind Skills](../skill-hub/create-and-bind-skills)
* [Conversation Prompts](../agent-settings/conversation-prompts)
* [Manage Agent Lifecycle](../agent-settings/agent-settings)