Automating cross-repo documentation with GitHub Agentic Workflows

· Source: The GitHub Blog · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering, Cybersecurity & Data Privacy · Depth: Intermediate, medium

Summary

The Aspire team successfully implemented GitHub Agentic Workflows to automate cross-repository documentation, addressing the challenge of securely managing docs in `microsoft/aspire.dev` for code in `microsoft/aspire`. This system enabled 82 feature-docs pull requests to merge at a median of 44.8 hours after their corresponding product pull requests, with 100% of created drafts being merged. The workflow, triggered by merged product PRs, uses an AI agent to draft documentation based on code changes and linked issues. A critical security feature is the "safe-outputs handler," which materializes the agent's intent (e.g., creating a pull request) using a narrowly scoped GitHub App token, restricted to specific repositories and actions. This ensures the agent has read-only access, while writes are explicitly allowed and reviewed by the feature's original engineer. The approach significantly reduced the "reverse-engineering tax" on docs writers and improved documentation timeliness.

Key takeaway

For DevOps or MLOps Engineers managing documentation across separate code and docs repositories, especially within strict security environments, you should evaluate GitHub Agentic Workflows. This system automates draft documentation creation and assigns the original feature engineer for review, significantly reducing documentation lag and manual effort. Consider starting with a single workflow like `pr-docs-check` to observe its impact on your team's median time-to-docs and free up your technical writers for conceptual work.

Key insights

GitHub Agentic Workflows automate cross-repo documentation securely by using AI agents with constrained write access and human review.

Principles

Method

GitHub Agentic Workflows use a markdown-defined agent to process code changes, generate documentation intent, and then a "safe-outputs handler" creates draft pull requests in a separate docs repository, assigning the feature's SME as reviewer.

In practice

Topics

Best for: AI Engineer, MLOps Engineer, DevOps Engineer

Related on AIssential

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by The GitHub Blog.