ValidMind Release Notes

Content Engineering

I automated our fragmented and manual release notes process using Python scripts and use of agent skills supported by large language models, reducing writer input time from almost a week to less than a day.

Situation

At the end of my tenure at ValidMind, the volume of our release notes required a writer to spend up to a week:

• Manually collating and verifying information on features for multiple products in our offering from many competing sources, including cloud storage, project management boards, and scattered GitHub repositories.
• Creating multiple versions of release notes for each product in our suite, one for public consumption, and one tailored for on-prem customers.

Solution

Using Cursor Code Agent Skills¹ supported by large language models (LLMs) and custom Python scripts, I automated collating and drafting release notes for all our products:

¹ Cursor: Agent Skills

1. When a release tag is cut on relevant repos, a GitHub Workflow fires to detect the delta between the previously published set of release notes and the new tag.

   Any missing release notes are then drafted using LLMs according to logic per release type (on-prem patch, python library, comprehensive monthly product suite release) from the feature pull requests (PRs) falling within that new tag’s release scope, guided by a skill informed by release notes conventions and our company documentation style guide.

2. Drafts for relevant release notes versions for that release tag’s scope are generated in our release-notes repository in a branch and a PR is automatically opened when the generations are complete.

   • This branch is linked automatically to a ticket cut in our project management board for editing the release notes for that tag’s scope.
   • A notification is sent to the release-notes Slack channel² alerting that release notes are ready for review and editing.

3. PR comment bodies contain information about the tag’s scope, and instructions for the human editor on how to wrap up and finishing curating the release notes.

   Release content that is served in multiple versions is single-sourced,³ meaning that editors only need to review one file for content displayed in multiple versions instead of many files.

² Slack

³ Portfolio Framework

The custom Python scripts also enable authors to generate LLM-drafted release notes manually if the automation is not available due to GitHub Workflows outages or other issues.

I also delivered comprehensive internal documentation for the automation to empower authors to review and edit the generated release notes in accordance with our standards.

Citation

While the solution I engineered was proprietary and cannot be shared publicly, I was permitted to showcase modified versions of the documentation I created to capture my automation’s functionality and the standard operating procedures users were expected to follow.

You can some of these artifacts with only the most relevant guides included with references to proprietary content or repositories removed using the portal below:

ValidMind Release Notes Guides