Skip to content

Latest commit

 

History

History
248 lines (179 loc) · 6.25 KB

create-a-new-task.md

File metadata and controls

248 lines (179 loc) · 6.25 KB

Create a new Docs Agent task

This guide provides instructions on how to create a new Docs Agent task file (.yaml) for automating a workflow.

Create a new task

To create a new task in your Docs Agent setup, follow these steps:

  1. Go to your Docs Agent directory, for example:

    cd ~/docs_agent

  2. Go to the tasks directory:

    cd tasks

  3. Open a text editor and create a new yaml file, for example:

    nano my-new-task.yaml

  4. Copy and paste the following task template:

     tasks:
   - name: "<TASKNAME>"
     model: "models/gemini-1.5-flash"
     description: "<DESCRIPTION>"
     preamble: <”PREAMBLE>”
     steps:
       - prompt: "<PROMPT_1>"
       - prompt: "<PROMPT_2>"
       - prompt: "<PROMPT_3>"

  5. Update the task name (<TASKNAME>).

  6. (Optional) Update the model (see this Gemini model code).

  7. Update the task description (<DESCRIPTION>).

  8. (Optional) Update the preamble (<PREAMBLE>).

    Note: The preamble field is not required. If it's not used, remove the preamble field.

  9. Update the prompts (<PROMPT_#>).

  10. (Optional) Add more prompts under the steps section.

  11. Save the file and exit the text editor.

  12. To verify that your new task is available, run the following command:

    agent runtask

  13. Run the new task using the task name:

    agent runtask --task <TASKNAME>

    For example:

    agent runtask --task MyNewExampleTask

    If your task accepts custom input, you can run it with the --custom_input field, for example:

    agent runtask --task MyNewExampleTask --custom_input ~/my_project/my_example_doc.md

A minimum task file

The example task below is created using only the required fields:

tasks:
  - name: "MyExampleTask"
    model: "models/gemini-1.5-flash"
    description: "An agent example that is created using only the required fields for a task."
    steps:
      - prompt: "This is my first prompt."
      - prompt: "This is my second prompt."
      - prompt: "This is my third prompt."

A task can have one prompt step at a minimum.

For more examples, see the tasks directory.

More examples for steps

This section contains more prompt examples and optional fields.

A standard prompt step

A step that runs the helpme command:

   steps:
     - prompt: "Revise the PSA email draft above to be more concise and better structured."

A POSIX command step

A step that runs a POSIX command:

   steps:
     - prompt: "git --no-pager log --since=2024-10-15"
       function: "posix"

Important: To run a POSIX command, the function field must be set to posix.

A step that reads a file

The file flag reads the specified file and added its content to the prompt's context.

A step that runs the helpme command with the file flag:

   steps:
     - prompt: "Analyze this file to understand the overall structure and key concepts."
       flags:
         file: "/home/user/docs_agent/README.md"

A step that runs the helpme command with the file flag and accepts custom input:

   steps:
     - prompt: "Analyze this file to understand the overall structure and key concepts."
       flags:
         file: "<INPUT>"
         default_input: "/home/user/docs_agent/README.md"

When this step is run, the <INPUT> string will be replaced with the value provided in the --custom_input field at runtime.

A step that reads all files in a directory

The allfiles flag reads all the files in the specified directory and added their content to the prompt's context.

A step that runs the helpme command with the allfiles flag:

   steps:
     - prompt: "Analyze the files in this directory to understand the overall structure and key concepts."
       flags:
         allfiles: "/home/user/docs_agent/docs"

A step that runs the helpme command with the allfiles flag and accepts custom input:

   steps:
     - prompt: "Analyze the files in this directory to understand the overall structure and key concepts."
       flags:
         allfiles: "<INPUT>"
         default_input: "/home/user/docs_agent/docs"

When this step is run, the <INPUT> string will be replaced with the value provided in the --custom_input field at runtime.

A step that reads each file in a directory

Similar to a foreach loop, the perfile flag reads each file in the specified directory and added the file's content to the prompt's context. For instance, if there are 5 files in the input directory, the same prompt will run 5 times using each file's content as context.

A step that runs the helpme command with the perfile flag:

   steps:
     - prompt: "Analyze this file to understand the overall structure and key concepts."
       flags:
         perfile: "/home/user/docs_agent/docs"

A step that runs the helpme command with the perfile flag and accepts custom input:

   steps:
     - prompt: "Analyze this file to understand the overall structure and key concepts."
       flags:
         perfile: "<INPUT>"
         default_input: "/home/user/docs_agent/docs"

When this step is run, the <INPUT> string will be replaced with the value provided in the --custom_input field at runtime.

A step with the name field

A step that runs the helpme command and the name field (which is for display only) is provided:

   steps:
     - name: "Revise the PSA email"
       prompt: "Revise the PSA email draft above to be more concise and better structured."

A step with the function field

A step that runs the helpme command that explicitly mentions which function to use:

   steps:
     - prompt: "Revise the PSA email draft above to be more concise and better structured."
       function: "helpme"

Without the function field, the prompt is set to use the helpme command by default.

A question step

A step that runs the tellme command:

   steps:
     - prompt: "Does Flutter support Material Design 3?"
       function: "tellme"

Using the tellme command requires a vector database setup.