---

name: chat description: "Use this RAGs sub-skill to query generated RAG agents, inspect text and image sources, and troubleshoot Generated RAG Agent chat behavior." disable-model-invocation: true metadata: disco-role: operating license: MIT

RAGs Chat

Use this sub-skill after a RAGs bot has been created and the task is to ask questions, debug generated-agent responses, inspect cited sources, or reason about text and image source display.

Read This When

• The user mentions the Generated RAG Agent page, chat UI, agent.chat, source nodes, image sources, or missing citations.
• The app says the agent is not created even though the user expected one.
• The generated answer is wrong, has no sources, has broken image paths, or appears to ignore summarization/retrieval settings.
• The task involves get_image_and_text_nodes, MultimodalChatEngine, or the response source_nodes contract.

For setup, data loading, or model selection before an agent exists, read ../builder/SKILL.md. For changing an existing agent's configuration or cache, read ../configuration/SKILL.md.

Core Workflow

1. Confirm current_state.cache.agent exists. If not, route the user back to the builder flow.
2. Ask the question through the generated agent chat route. The app appends the user message, calls agent.chat(str(prompt)), displays the response, and stores the response object in message metadata.
3. Inspect sources using the response's source_nodes when available. Text sources are displayed as IDs and full metadata content; image sources are displayed from metadata["file_path"].
4. If source evidence is absent or weak, debug retrieval/data/index settings before treating it as a UI problem.
5. For model/provider errors, route back to builder/configuration because the chat page uses the agent created from those settings.

Read references/chat-workflow.md for detailed query and source-display flow. Read references/api-reference.md for response and helper behavior. Read references/troubleshooting.md for known failure symptoms.

Important Operating Facts

• The chat page initializes agent_messages with "Ask me a question!".
• It only enables chat when the selected cache has a live generated agent.
• display_sources splits response.source_nodes into image and text nodes.
• Image nodes need a local file_path metadata value; text nodes use MetadataMode.ALL to include metadata in displayed content.
• Default agent construction creates a vector search tool and may add a summarization tool when include_summarization=True.
• External model calls and URL/web-search behavior require credentials and network access; the safe skill verification did not execute them.

Boundaries

Do not promise that a source list exists for every response. Some chat engines or error paths may not provide source_nodes, and poor retrieval settings can produce empty or irrelevant sources. Debug the data/index/config route before rewriting chat UI behavior.