How we used FalkorDB, Graph Theory (WCC, Betweenness, PageRank), and Multi-Agent simulation to catch breaking ecosystem changes before merge.

On October 16, 2026, Home Assistant merged a refactor that moved sensor entity classes into a separate module. While this change passed core tests and linting, 32 custom integrations failed to load due to internal symbols no longer being exposed. This scenario exemplifies Hyrum's Law, where depending on a system's behavior can lead to unintended issues for downstream users, especially with the rise of AI coding agents that quickly refactor code.

To address this, developers introduced ShadowGraph, a tool developed during the WeMakeDevs × FalkorDB Graph Hacks Hackathon. ShadowGraph helps developers anticipate the impact of a code change before merging it. It builds a dependency graph using Python AST analysis to identify which integrations rely on which symbols. By creating an isolated copy of the graph, ShadowGraph can simulate changes to see what would break if a symbol were removed or altered, distinguishing between immediate import failures and potential runtime issues.

The tool leverages AI agents to verify potential breakages and compares findings against static analysis, suggesting migration fixes to maintainers. ShadowGraph represents the Home Assistant dataset, containing 4,814 integrations and 562,444 import relationships, efficiently exploring how a change can affect downstream projects. Its graph-copy capability allows for isolated simulations without altering the live graph.

ShadowGraph’s approach extends beyond traditional CI, which only checks if your own tests pass. It asks the critical question: “Your tests pass, but what breaks for everyone depending on your code?” ShadowGraph is open source under the MIT License and is built for Graph Hacks across three tracks: connected data, agent memory and coordination, and company knowledge. The ultimate goal is to catch breaking changes before users experience them.