Defining and measuring quality across diverse agent platforms is essential as the line between developer-written code and agent-generated code blurs. Currently, agent-specific guardrails lack cross-compatibility.

This session uses an AGENTS.md file which lives in memory for multiple chat sessions and acts as a universal agent README, providing context, instructions, and references to integrated static analysis tools (Check style, PMD, Spot bugs). This is a case study presentation that includes a live demonstration of the pipeline and quantitative data analysis.

The demonstration shows initial agent-generated code failing static analysis checks. Subsequent generation, after introducing the `AGENTS.md` file, results in an approximate 85% reduction in violations and 60% increase in unit test coverage. Attendees learn to craft `AGENTS.md` to ensure enterprise-ready, standards-compliant Java code and leverage static tools as verifiable quality guardrails.

Presented by Shuchita Prasad (J P Morgan) at JavaOne 2026 (CA, March 2026).

All https://www.youtube.com/playlist?list=PLX8CzqL3ArzUMVSzm-z_-if8BIB55EGl4 talks.

➠ https://github.com/ShuchitaTechTalks/agent-agnostic-guardrails
➠ https://agents.md
 
Tags: #Java #JavaOne #JDK

LRN1175 17 104

Agent-Agnostic Guardrails: Universal Java Code Quality with AGENTS.MD and Static Analysis