tools-only Instructions Example
This document demonstrates how to use the instructions parameter to control LLM behavior with system prompts.
Source Content
{{ ref('source.md').content }}
Without Instructions
Summary: {{ ref('source.md') | llm_extract('a brief one-sentence summary of the feedback') }}
With Pirate Instructions
Pirate Summary: {{ ref('source.md') | llm_extract('a brief one-sentence summary of the feedback', instructions='Talk like a pirate. Use "Arrr" and pirate vocabulary.') }}
LLM Block with Instructions
{% llm instructions='Talk like a pirate. Use "Arrr" and pirate vocabulary.' %} Summarize this product feedback in one sentence: {{ ref('source.md').content }} {% endllm %}
Using Classification from Another Model
{% llm instructions='You are a customer success manager.' %} Based on this sentiment analysis:
{{ ref('classify.md').content }}
Write a one-sentence professional response acknowledging the feedback. {% endllm %}