AI Agents Debugger Introduced for ReACT Workflows in Drupal

The tool was developed by Marcus Johansson, Tech Lead of the Drupal AI Initiative and Senior Developer at FreelyGive. According to his announcement, the project responds to practical challenges observed during hackathon sessions, where developers relied on general-purpose tools such as Xdebug to trace agent behaviour.

The debugger is designed around the internal logic of ReACT (Reason + Act) agents introduced in AI Agents 1.1.0 — a reasoning pattern in which agents alternate between analytical steps and tool execution. The term refers to this agent architecture and is unrelated to the JavaScript library React. By aligning the debugger with this reasoning model, the tool aims to provide visibility that traditional debugging utilities cannot offer.

It records each execution step, surfaces agent-to-agent conversations, and allows import and export of run data for review or replication. Early community feedback has already led to enhancements, including clearer aggregation of token usage and runtime visibility per agent call.

Unlike earlier workflow inspection approaches, the module focuses on native observability for AI-driven processes. Developers can review reasoning loops, inspect intermediate states, and analyse how sub-agents interact during complex tasks. The project is currently marked as beta, with a demonstration video and installation instructions available on its Drupal.org project page.