Skip to content

Add more info to system prompt to allow for the tools to process the user provided prompts in a more orderly way. #127

Open
@sangee2004

Description

@sangee2004

When i need the tool to process a set of steps , it is not enough that i state the steps in an order like this example:

Tools:  github.com/gptscript-ai/vision,  github.com/gptscript-ai/image-generation , sys.write, sys.read, sys.download

1. Generate an image of a cow grazing in the open green meadows and return only the url of the image.
2. Download the url of the image generated in step 1 and write to a file cow.png.
3. Describe the image at cow.png in the style of a Wes Anderson character.

When i run the above script , i always end up getting the following error:

10:59:00 sent     [main]
10:59:00 started  [sys.read(2)] [input={"filename": "cow.png"}]
10:59:00 sent     [sys.read(2)]
10:59:00 started  [https://raw.githubusercontent.com/gptscript-ai/image-generation/b9d9ed60c25da7c0e01d504a7219d1c6e460fe80/tool.gpt(3)] [input={"prompt": "a cow grazing in the open green meadows", "quality": "hd"}]
10:59:00 sent     [https://raw.githubusercontent.com/gptscript-ai/image-generation/b9d9ed60c25da7c0e01d504a7219d1c6e460fe80/tool.gpt(3)]
10:59:00 failed to run tool [] cmd [/usr/bin/env python3 /Users/sangeethahariharan/Library/Caches/gptscript/repos/b9d9ed60c25da7c0e01d504a7219d1c6e460fe80/python3.12/cli.py --prompt=a cow grazing in the open green meadows --size= --quality=hd --number=]: context canceled
2024/03/07 10:59:00 open cow.png: no such file or directory

When i include the following lines additionally in prompt - Follow all steps in order and move to the next step only after the first step is done completely. , it works as expected.

Tools:  github.com/gptscript-ai/vision,  github.com/gptscript-ai/image-generation , sys.write, sys.read, sys.download

Follow all steps in order and move to the next step only after the first step is done completely.

1. Generate an image of a cow grazing in the open green meadows and return only the url of the image.
2. Download the url of the image generated in step 1 and write to a file cow.png.
3. Describe the image at cow.png in the style of a Wes Anderson character.

Should we consider adding something along the lines of Follow all steps in order and move to the next step only after the first step is done completely. to our existing system prompt ?

Metadata

Metadata

Assignees

No one assigned

    Labels

    proposalAn enhancement or documentation proposal

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions