PMM-7-Cross AI Agents optimization

[ondrejpatocka]

Change Draft mode

• this is to sparkle discussion and team experimentation in this subject :)
• PMM consists of 14 repos. next steps would be *.md coverage and cross linking of those components/repos
• validation, finalization, adoption and continuous maintenance of these files is subject to further discussion

Summary

• Add AGENTS.md as the single authoritative entry point for AI coding assistants (Claude Code, Cursor, GitHub Copilot, and others), covering PMM product overview, architecture, data flow pipelines, domain model, tech stack, global conventions, and cross-links to all component guides
• Add cross-agent compatibility shims so all major AI assistants auto-discover the documentation:
  • CLAUDE.md — Claude Code entry point
  • .cursorrules — Cursor entry point
  • .github/copilot-instructions.md — GitHub Copilot entry point

Motivation

LLM-based coding agents (Cursor, Claude Code, Copilot) work significantly better when they have structured context about whole codebase's architecture, domain model, patterns, and conventions. PMM is a large multi-component system spanning Go backends, a React frontend, protobuf APIs, ClickHouse, VictoriaMetrics, multiple QA repos and complex build infrastructure.

Without guardrails and guidance, agents frequently:

• Hallucinate
• Spend more tokens
• Miss important parts

These docs give every agent immediate, scoped context for the product, repo and component being worked on.