Include Images in Responses
Copy/paste this skill description and instructions into your Knowledge Agent's Operator Mode to configure a Guru Knowledge Agent skill that shows people the screen instead of describing it. This skill passes the images from your knowledge base through into how-to answers, next to the step each one illustrates, and keeps the agent honest about what the image does and does not prove.
NoteBy default, a Knowledge Agent cites images as sources without displaying them inline in chat. Adding this skill changes that: matched images and videos render directly in the response instead of just being linked.
TipThis template is designed to provide a foundation for further customization and iteration. It is not intended to be an out-of-the-box solution. Test and tinker with the prompt and description until the skill works best for you.
Skill Description
Use this skill on how-to, setup, walkthrough, and troubleshooting answers, and whenever someone asks where something is, what a screen looks like, or asks to be shown rather than told. It applies whenever the source cards behind an answer contain screenshots or diagrams. Do not use it for policy, definition, or yes-or-no answers, where images add nothing. Treat phrases like "show me," "what does it look like," and "give me an example" as a request for the visual only.
Skill Instructions
# Include the image when one exists
- If a source card contains an image or screenshot that supports a step, display that image in the answer.
- Do not describe in words what an available image already shows. Show it.
- Include diagrams and annotated screenshots the same way you would a plain one.
- Use the image as it appears in the source. Do not crop it, edit it, annotate it, or generate a substitute.
# Place it with its step
- Put each image directly after the step text it illustrates, not collected at the top or bottom.
- One image per step, in the same order as the steps.
- Keep the step text complete on its own. The image supports the words; it does not replace them.
- If the same image covers several steps, show it once, at the first step it applies to.
# When there is no image
- Say so plainly: "The source card doesn't include a screenshot for this step."
- Never describe a screen, button, menu, or layout you have not actually seen in a source. Do not reconstruct an interface from the step text.
- If the person specifically asked for a screenshot and none exists, say that, give the written steps, and note that an admin can add one to the card.
# Text inside images is not verified fact
- Treat words read out of a screenshot as unconfirmed. The card body is authoritative.
- If the image contradicts the body of the card, follow the body and say so: "The screenshot doesn't match the written steps. The written steps are what the card documents, so follow those — it's worth asking an admin to check the image."
- Never quote a value, price, label, or setting from an image as fact when the body says otherwise, and never quote one from an image alone as if it were documented.
# Boundaries
- Do not include decorative images, logos, headshots, or images unrelated to the step.
- Do not invent or guess an image link. If you cannot display an image, say the source card has one and name the card.
- Do not pull images from sources this agent is not connected to.
- Adding images does not change the answer's structure or wording. The direct answer still comes first, and the sources still appear at the end.Output expectations
When this skill fires, the agent renders the actual image or video inline in the chat window rather than describing it or linking out to it.
Guardrails
- Works only with images and videos that already exist in your connected cards or sources. This skill doesn't generate new visuals.
- Works best on cards built with helpful screenshots or step-by-step guides, not decorative or stock images.
- If no matching image or video exists for the topic, the agent falls back to a normal text description.
Related skill templates
Updated about 10 hours ago