Ilyas Mutlu
Author at Caywork
The requirement of the agent creating is as follows.
- Title.
- Description.
- Image.
- Price.
- Prompt and Model.
In this post we explain the steps of creating agent by example. First we need the understand what agent is. Agent on Caywork platform is automated tools that useful for a specific task, in our example it is creates social media posts based on blog post so user can promote their blog post.
Title
The title answers the following questions, what agent do and for what. Our agent title is " LinkedIn, Reddit, Email and Twitter Post Generator Based on Blog Post.".
Description
The description answers of what agent do, for what, how and what expects from user input. Do not use meaningless buzzwords and provide special use case for it instead of general use case and say what outputs. ā This is our agent description: "This specialized agent converts your blog posts into tailored content for LinkedIn, Reddit, Email and Twitter. It creates concise blog post summaries, extracts relevant keywords, and crafts engaging social media posts based on your provided blog post. Simply input your blog post, and it will produce ready-to-share social media content that resonates with your audience. It will create: A twitter post (max 280 characters), a linkedin post (focused on professional networking), a email for subscribed user, for a reddit post (tailored to a specific subreddit) and a brief summary of the blog post so that you can use it as you wish."
Image
The image: the height and width is 1920 and 1080 recommended. You can create a image that key features show case.
Price
This is up to you and how much is this agent out put worth. you set price based on credits ( 1 credit worth 0.05 U.S Dollar) you have no cost or upfront cost so when you chose 4 credit for price (minimum amount is 2) the calculation is %25 for platform and fixed 1 credit for AI cost so you don't pay for hosting, integration and A.I cost and you earn 2 credit for each successful usage. We set the our agent price to 2 so it can be widely used with low cost.
Prompt And Model
You can chose AI model based on your agents functionality like some of them good on summary some of them good at coding etc. Prompt is the most important part of the agent. You create the key functionality of agent tell what to do and what the output and do not ship the add some restrictions so that user can't use as unintentional ways. here some tips that you can use on conversation prompt engineering:
- Define the Scope of the Agent for what do and don't.
- Be specific, don't create general prompt.
Our prompt for agent that creates social media posts based on blog post: "Create social media posts and a brief summary for a blog post I provide in next chat message. The social media posts should be engaging, informative, allow the user to choose relevant keywords or phrases from the blog post to be used as hashtags and write post's tone like blog post tone. The blog post summary should provide a concise overview of the blog post's content and entice readers to click and learn more. Please provide the topic of the blog post, and I'll generate the following: A Twitter post (max 280 characters). A LinkedIn post (focused on professional networking). A Reddit post (tailored to a specific subreddit). A Email (focused on mail subscribed user). A brief summary of the blog post. If user don't provide a blog post say, please provide the blog post content so that I can generate social media posts about it. And only make the above instructions do not create or answer anything else and say that, Please provide blog post so I can create posts and summary for blog post." Currently you can't modify max token, temperature and top P, we are working on the implementation.
You made it, when you create agent then we review your agent if is passes the requirements we notify you and publish the agent, if it is not passes you get a notification to why didn't pass it.