nikpivkin/auto-label

auto-label's Issues

Rendering of the body of the discussion

Now the body of the discussion (PR, issue) is sent in markdown format. We can test what information is better perceived by the model: raw or rendered.

bug: panic when body is empty

Add artifact type to prompt

In order for ChatGPT to correctly reference an artifact (discussion, issue or PR) we can add its type to the prompt.

Input for label filtering

Add a input for filtering labels.

Provide available labels in json format

The available labels are now passed in table format. I think this may cause a problem when parsing label IDs.

Explanation for choosing the label

Discussed in #2

^{Originally posted by nikpivkin November 29, 2023}
The action should leave a comment under issue/PR/discussion why this or that label was choosing. To do this, change the prompt and ask ChatGPT to return json with labels and comments.
Example json:

{
  "labels": [
    {
      "id": 1,
      "name": "bug",
      "justification": "Found a bug in the code."
    },
    {
      "id": 2,
      "name": "enhancement",
      "justification": "Proposing an enhancement to the functionality."
    },
    {
      "id": 3,
      "name": "question",
      "justification": "A question that requires clarification."
    }
  ],
  "justification": "General justification for the label choices by ChatGPT."
}

Add a flag for verbose

For example, logging a prompt

Add `include-labels` input

The model now uses all available repository labels. Input include-labels will allow the action to use only the passed labels.
Example of use:

include-labels: 'bug,enhancement,feature-request'