Editorial Automation: Why & How to Set Up Chat-Ops for your Own Review System on GitHub

[This article was first published on rOpenSci - open tools for open science, and kindly contributed to R-bloggers]. (You can report issue about the content on this page here)
Want to share your content on R-bloggers? click here if you have a blog, or here if you don't.

Anyone can contribute a software package to the rOpenSci suite as long as it fits our scope (research lifecycle software and statistical software) for a transparent, constructive, nonadversarial and open review. In practice, the review steps are all recorded in GitHub issue threads (example). Software peer-review involves coordinating and tracking many moving parts: software submissions (new issues), testing and diagnostics, assignment of editors and reviewers, and logging the progression of submissions through revisions and acceptance.

Our editorial workflow has been significantly enhanced with the use of ✨ automation ✨ .

To ensure that a package meets our compliance standards, we would previously clone the repository locally, install dependencies, run manual checks, gather the results, and copy-paste them into the issue thread. Now we can accomplish the same outcome with just a simple command into a GitHub issue comment:

@ropensci-review-bot check package

Similarly, we can use the following command to register a reviewer named in the submission issue metadata (filling YAML data) as well as in our Airtable database.

@ropensci-review-bot add @maelle to reviewers

Feeling inspired? Maybe you run a submission process (of papers, software, conference abstracts) that is handled in GitHub issue (or pull request) threads (or could be moved there)?

In that case, maybe you could benefit from the same toolset as us: chat-ops where a GitHub “bot” account performs actions dictated by human-typed commands, orchestrated by a Ruby app deployed on a cloud platform.

In this post, we explain how to assess whether that app framework, the editorial bot generator Buffy, might be right for you. We detail the steps involved in setting up “chat-ops” with Buffy for your own needs.

What is an editorial bot generator?

The editorial bot generator Buffy is the tool that supports the aforementioned commands. It is a Ruby codebase that can be

  • configured (and potentially extended),
  • then deployed on a cloud platform like Heroku where it is always on and listening,
  • and hooked into GitHub issues or pull requests via, well, webhooks. 🪝

Every time a new issue is opened, every time an issue comment is created, its contents are sent to the deployed app. If it corresponds to the regular expression of a registered command, predefined steps are launched: a check is launched somewhere, a database is updated, information is copied back to the GitHub issue, etc.

The “visible” bot is the GitHub account used as bot face: commands are addressed to it, and it uses a “Personal” Access to post GitHub issue comments.

The Journal of Open Source Software (JOSS) has developed an initial list of possible commands/actions, also known as responders. One can create new responders by writing Ruby code following documented instructions. Note that one of the responders allows launching a GitHub Action Workflow which can circumvent the absence of Ruby knowledge on a team – although a pure Ruby responder might be faster. You can also send a call to any API, so if you can build an external API, you can really implement many things.

The editorial bot generator is for you if…

  • You run a submission process (of papers, software, conference abstracts) that is handled in GitHub issue (or pull request) threads (or could be moved there)?
  • Your process involves tedious steps (editing issue/PR comments, switching issue/PR labels, copy-pasting URLs into an external database, running automatic checking tools) that can be automated via scripts possibly interacting with web APIs?
  • You can recognize interesting responders in Buffy docs or you have Ruby talent on your team or contractor contacts, who could write custom responders for you?
  • Do you have time and resources to spend at least a few days setting it up and communicating the change of processes to your users?
  • Can you devote some time to maintaining the installation e.g. responding to Heroku security updates or keeping Buffy codebase up to date with upstream changes?

They use the editorial bot generator

How to set up the editorial bot generator for your system

We shall first show how Buffy usage works, afterwards, we shall go into details about how to get there. The following diagram represents the whole automation toolset we use for rOpenSci software peer-review.

Diagram representing automation for rOpenSci software peer review. On the left, a GitHub issue thread with emojis as avatars, and wobbly lines as text. Under the GitHub issue thread, a legend indicating who among the emojis is Author /Editor / Reviewer / ropensci-review-bot. At the center of the diagram is an Heroku app using the Buffy Ruby tool, that receives information from GitHub via webhooks. The app digests messages received and depending on the command pings an external service represented on the right (with a plumber logo); fills the Airtable-based software review database; manages ropensci GitHub organization via GitHub API; posts back or labels in the GitHub issue thread.

We can observe, on the left, a GitHub issue thread corresponding to a submission. The submission has

  • a title,
  • a body that was created based on a GitHub issue template,
  • labels indicating the progress of the submission from editorial checks to the final decision,
  • an assignee who is the editor,
  • various commenters who are the author, the editor, reviewers, and the bot account.

Every time an issue is created or updated with a new comment, information about that event is sent to the central app via a webhook. There on Heroku, if the comment corresponds to a registered command, actions (the formerly tedious steps 🙂) are made accordingly. Possible actions include filling an Airtable base via Airtable API, launching an external software check on a plumber API, inviting the author to the ropensci GitHub organization, posting a comment back into the GitHub issue thread with some results or confirmation.

How does one achieve this?

Initial preparation & installation steps

Follow Buffy installation instructions.

  • Fork the Buffy codebase to an organization of yours, and create a branch there. Ours is named ropensci. The organization does not have to be where the review repository also lives.
  • Create a test review repository, that is to say, a copy of your production review repository so you can experiment without bothering serious watchers. The test repository should contain the same issue/PR templates and issue/PR labels as the production repository.
  • Create a bot account (save its credentials and 2FA method into, for instance, your team’s 1Password vault). Give it access to your production and test review repositories. It might even need more access to your GitHub organization based on what you’ll task it to do. Follow Buffy docs to create a Personal Access Token, save it temporarily on your computer as you’ll need to save it in the app configuration.
  • Set up a Heroku account and app for Buffy deployment – or do the same on another service such as Render. Following the instructions worked for us. Make sure your pricing tier allows for the app to listen all the time. If the app is sleeping it will not be able to digest comments from GitHub.
  • Check the build logs of your Heroku apps indicate success.
  • In your test and production repositories, set up a webhook to send GitHub issue/PR comments to Heroku or your other service.

As mentioned in the docs, at this stage in your test review repository you can write the following comment (replace the username with your bot account username)

@ropensci-review-bot help

What if it does not work?

  • Re-read the installation steps to ensure you did not miss anything.
  • Look into the webhooks of your repository, maybe there is a failure message there.
  • Consult the logs of the Heroku app (we found it most convenient to use Heroku CLI for this… to copy-paste info to our Ruby developer).

Configuration, tests, documentation

Now comes the time to adapt your Buffy version to your needs! Good news: you can keep doing this forever depending on how your needs evolve. Bad news: you will keep doing this forever as you’ll always see opportunities for improvement. 😉

To configure your Buffy installation you will be making changes in these places

  • In the /config/settings-production.yml file of the branch of your buffy fork;
  • In other folders of the branch of your buffy fork if you are adding custom responders;
  • In issue templates (.github/ISSUE_TEMPLATE) or PR templates (.github/PULL_REQUEST_TEMPLATE) and buffy templates .buffy/templates of your review repository (or repositories, if you created a test review repository for experimenting with Buffy, which we’d recommend). Indeed, issue or PR templates will contain placeholders/wrappers for HTML variables like <!--editor--> <!--end-editor--> – otherwise the bot won’t be able to fill this information. buffy templates are for comments you will want the bot to post, for instance, a checklist at the end of the review process.

Follow Buffy docs on configuration. You will be adding (registering) responders by adding them to the YAML file /config/settings-production.yml, with subfields indicating some options. For instance, you might want to use the “assign editor” responder to store the editor username in the issue comment without assigning the issue to them so you’ll set add_as_assignee to false.

You’ll find responders and their parameters in Buffy docs. You can also check out the readthedocs website of rOpenSci’s version of Buffy in case some of our custom responders are relevant for you (they are at the bottom of the list, with rOpenSci in front of their name).

After each responder addition or configuration, try it out by creating issues (or pull requests if that’s your process) and typing comments in them. If it works, you will be convinced you have added one feature to your system, congratulations!

Afterward, the feature should be officially released by telling actors of your system about it. In our case, we wrote announcements in the slack channel we have for editors, and we updated our dev guide. Updating guidance is particularly rewarding as bot commands typically replace lines of tedious task descriptions. 😁

Conclusion

In this post, we presented the editorial bot generator Buffy. We hope to make it easier for you to choose whether to adopt it for your own submission system and to know how to adopt it. The costs linked to Buffy usage are:

  • developer time to set it up, tweak or add responders, document its usage, and long-term maintenance;
  • users’ time to learn how to use GitHub comments (lower cost for newcomers to your system, higher cost for those who had gotten used to tedious steps);
  • a subscription to Heroku or a similar service to ensure the app is always listening.

In our experience, adopting Buffy has been worth it as once it’s well adopted, it

  • decreases the cognitive load needed for handling a review as one does not need to switch between different tabs or apps;
  • simplifies future process changes, as the command could remain the same whilst the background tasks change.

Feel free to comment with any questions you might have about Buffy!

To leave a comment for the author, please follow the link and comment on their blog: rOpenSci - open tools for open science.

R-bloggers.com offers daily e-mail updates about R news and tutorials about learning R and many other topics. Click here if you're looking to post or find an R/data-science job.
Want to share your content on R-bloggers? click here if you have a blog, or here if you don't.

Never miss an update!
Subscribe to R-bloggers to receive
e-mails with the latest R posts.
(You will not see this message again.)

Click here to close (This popup will not appear again)