Dupliket

Dupliket is a GitHub action that helps repository maintainers to triage issues and discussions efficiently by finding duplicates and reference them directly in the newly created issue or discussion. It is powered by LLM provided by OpenAI to process issues and calculate their similarites.

Installation

You can integrate Dupliket in your GitHub workflow by creating a new workflow file, for example:

on:
  issues:
    types: [opened]

jobs:
  dupliket:
    runs-on: ubuntu-latest
    steps:
      - name: Find possible duplicates
        uses: Namchee/dupliket@<version>
        with:
          access_token: ${{ secrets.GITHUB_TOKEN }}
          api_key: ${{ secrets.API_KEY }}

Supported Events

Dupliket supports the following workflow events:

Event	Types	Description
`issues`	`created`	Fired whenever a new issue is created.
`discussion`	`opened`	Fired whenever a new discussion is created.

Please refer to the workflow events reference for more information about the supported events.

Since GitHub includes pull request on issues events, it is recommended to filter pull request events when listening to issues events. You can filter pull request using job conditions in your workflow file. For example:

on:
 issues:
   types: [opened]

jobs:
 dupliket:
   runs-on: ubuntu-latest
   if: ${{ !github.event.issue.pull_request }} # Do not listen to pull request events
   steps:
     - name: Find possible duplicates
       uses: Namchee/dupliket@<version>
       with:
         access_token: ${{ secrets.GITHUB_TOKEN }}
         api_key: ${{ secrets.API_KEY }}

Inputs

Dupliket accepts the following input that can be filled in jobs.with section:

Name	Required?	Default Value	Description
`access_token`	`true`	This field is required	GitHub's access token. Used to interact with GitHub API for managing knowledge and creating comments. Note: Ensure that your token have permissions to read and write code and issues
`api_key`	`true`	This field is required	OpenAI's API key. You can get it by signing up for an OpenAI account
`model`	`false`	`text-embedding-ada-002`	Language model to be used when searching and calculating issue and discussion similarity.
`max_issues`	`false`	`3`	Maximum number of possibly similar issues and discussions to be displayed
`min_similarity`	`false`	`0.9`	Minimum similarity for an issue or discussion to be considered as similar. Must be a floating point between `0.0` and `1.0`
`show_similarity`	`false`	`false`	Include similarity percentage as footnote
`discussions`	`false`	`true`	Include discussions when searching for similar references
`label`	`false`		Label to be applied when an issue or discussion has duplicates. Fill with an empty string to not apply any labels
`template`	`false`		Apply a custom message when an issue or discussion has duplicates. See section customizing message for more detailed information on how to use this field.

Customizing Message

By default, dupliket will create a message containing references existing issues and discussions when an issue or discussion is created and has potential duplicates. While this is good enough for normal use-cases, you might want to change the message to suit your needs or want to change the formatting instead.

You can customize the message by providing a mustache template to template option in the workflow file. Markdown formatting is supported.

Below are the list of replacable values that you can use in your custom message:

Name	Description
`user`	Username that triggers the event
`count`	Number of similar issues and discussions
`references`	List of issue and discussion links

For example, given the below template:

Hi {{ user }},

Thanks for reporting an issue! However, it seems like there are {{ count }} issue(s) that are similar to yours:

{{ references }}

the action may reply with the following comment

Hi @Namchee,

Thanks for reporting an issue! However, it seems like there are 3 issue(s) that are similar to yours:

- https://github.com/Namchee/dupliket/issues/49
- https://github.com/Namchee/dupliket/issues/48

License

This project is licensed under the MIT License

Namchee / dupliket