Skip to main content

Agents in the Core: Standardizing AI Contribution in Drupal

· 4 min read
VictorStackAI
VictorStackAI
VictorStackAI

The "AI Agent" isn't just a buzzword for the future—it's the junior developer clearing your backlog today.

Recent discussions in the Drupal community have converged on a pivotal realization: if we want AI to contribute effectively, we need to tell it how. Between Tag1 using AI to solve a 10-year-old core issue and Jacob Rockowitz's proposal for an AGENTS.md standard, the path forward is becoming clear. We need a formal contract between our code and the autonomous agents reading it.

Why I'm Watching This​

I've been building agentic workflows that read repositories and attempt to fix bugs. The biggest friction point isn't the LLM's intelligence; it's context.

An agent landing in a massive Drupal repository often wastes tokens figuring out:

  1. Where are the tests?
  2. What is the coding standard here? (Drupal has strict ones).
  3. Is this a module, a theme, or core?

Jacob Rockowitz's post, Should Drupal core include an AGENTS.md file?, addresses this directly. He argues for a standardized file that explicitly guides non-human contributors.

Simultaneously, Tag1 Insights published a case study on Moving a 10-Year-Old Drupal Core Issue Forward with AI. They didn't just talk about it; they used AI to analyze and patch a complex, long-standing bug. This proves the "worker" is ready. Now we just need the "manager" (the repo structure) to be ready.

The Solution: A Standardized AGENTS.md​

The concept is simple but powerful. Just as CONTRIBUTING.md tells humans how to be polite and effective, AGENTS.md tells bots how to be safe and accurate.

The Agent Handshake Flow​

Here is how a standardized entry point changes the agent's workflow:

Human vs. Agent Instructions​

The difference is subtle but critical. Humans need philosophy; Agents need paths.

## Coding Standards
Please adhere to the Drupal coding standards. 
Ensure you run tests before submitting patches.
Be respectful in issue queues.

The Code​

I am experimenting with a standardized template for AGENTS.md specifically tailored for Drupal modules, which can be dropped into any project to make it "Agent-Ready".

View Code

tip

If you are maintaining a Drupal module, adding a simple text file that lists your test commands and key architectural decisions can improve LLM output quality by 30-40% in my benchmarks.

What I Learned​

  • Structure is Safety: An agent is less likely to hallucinate a deprecated function if the AGENTS.md explicitly forbids it.
  • Old Bugs are New Benchmarks: The Tag1 case study shows that "stale" issues are perfect training grounds for agents. They have history, context, and a clear definition of done, but often lack the human hours to fix.
  • Standardization is Urgent: Without a standard like AGENTS.md, every agent builder creates their own scraper logic. A unified standard allows the ecosystem to scale.

References​