Prompt engineering checklist

Check box	Check	Required or Optional	Notes
❑	Note that the LLM is incapable of web-scraping from a provided URL such that including these URLs in your prompt will not support the LLM in extracting any information.	Required
❑	Ensure all context provided in your prompt is useful to the LLM, especially when leveraging training data.	Required	Providing prompt context such as “Only answer if you know. If you don’t know, respond with ‘I don’t know.'” does not result in more accurate results.
❑	Be specific with your desired output instead of providing a simple command. For example, Summarize, Generate, Convert.	Required	If you want the LLM to summarize a body of text, elaborate on what specific terms or concepts the LLM should prioritize.
❑	Modify your API payloads to only include the required information, omitting any details that are not required for the task.	Required
❑	Define the LLM’s role in completing a task.	Required	For example, providing You are a text extraction agent will guide the LLM in providing more meaningful results.
❑	Dissect complex prompts to be composed of smaller sub-tasks.	Required	It is beneficial to support the LLM’s chain of thought by establishing smaller outputs that contribute to the final solution.
❑	Ensure that token limits will not impact the completion of your LLM response.	Required	Large prompts that consume a high number of tokens may result in the LLM’s response to return incomplete.
❑	Document all related Tools within a single Toolkit.	Required	Toolkits group multiple Tools and assist in coordinating Tool invocation for prompts that requires more than one Tool.
❑	Do not rely on the LLM to perform mathematical calculations.	Required	Depending on the model, even simple calculations may provide incorrect outputs and/or cause confusion.
❑	Review API documentation within a Tool for clear formatting.	Optional	Utilize the Tool Instructions for API documentation purposes.
❑	Utilize the temperature parameter to tailor the LLM response to your intended goal.	Optional	Lower temperature will result in concise and factual responses, while higher temperature will produce creative and diverse results.

❑

Note that the LLM is incapable of web-scraping from a provided URL such that including these URLs in your prompt will not support the LLM in extracting any information.

Required

❑

Ensure all context provided in your prompt is useful to the LLM, especially when leveraging training data.

Required

Providing prompt context such as “Only answer if you know. If you don’t know, respond with ‘I don’t know.'” does not result in more accurate results.

❑

Be specific with your desired output instead of providing a simple command. For example, Summarize, Generate, Convert.

Required

If you want the LLM to summarize a body of text, elaborate on what specific terms or concepts the LLM should prioritize.

❑

Modify your API payloads to only include the required information, omitting any details that are not required for the task.

Required

❑

Define the LLM’s role in completing a task.

Required

For example, providing You are a text extraction agent will guide the LLM in providing more meaningful results.

❑

Dissect complex prompts to be composed of smaller sub-tasks.

Required

It is beneficial to support the LLM’s chain of thought by establishing smaller outputs that contribute to the final solution.

❑

Ensure that token limits will not impact the completion of your LLM response.

Required

Large prompts that consume a high number of tokens may result in the LLM’s response to return incomplete.

❑

Document all related Tools within a single Toolkit.

Required

Toolkits group multiple Tools and assist in coordinating Tool invocation for prompts that requires more than one Tool.

❑

Do not rely on the LLM to perform mathematical calculations.

Required

Depending on the model, even simple calculations may provide incorrect outputs and/or cause confusion.

❑

Review API documentation within a Tool for clear formatting.

Optional

Utilize the Tool Instructions for API documentation purposes.

❑

Utilize the temperature parameter to tailor the LLM response to your intended goal.

Optional

Lower temperature will result in concise and factual responses, while higher temperature will produce creative and diverse results.