Grok Build CLI、CLAUDE.mdは読んでもhooksは検出しなかったTesting revealed that Grok Build CLI reads CLAUDE.md configuration files but…

A recent hands-on test reported on Qiita examined how xAI's Grok Build CLI handles configuration files originally associated with Anthropic's Claude Code, and found a partial compatibility picture: the tool reads CLAUDE.md context files but does not appear to detect or execute Claude hooks. For developers who work across multiple AI coding assistants, the finding matters because it clarifies exactly where interoperability holds and where it breaks down, helping teams avoid silent failures when they move a project between tools.

CLAUDE.md has become a de facto convention for storing project-level guidance that an AI coding agent should read at the start of a session. It typically contains instructions such as coding standards, build and test commands, directory structure notes, and conventions the model should follow. Because it is plain Markdown placed at the repository root, it is relatively easy for other tools to adopt. The test indicates that Grok Build CLI honors this pattern, loading the file's contents into context so that the assistant can align its behavior with the project's stated rules. That is a meaningful compatibility win, since it means much of the human-authored guidance carries over without modification.

Hooks, however, are a different mechanism. In Claude Code, hooks are configured callbacks that fire at defined points in the agent's lifecycle, for example before or after a tool call, or when a session starts. They are commonly used to run shell commands, enforce guardrails, trigger formatting or linting, or log activity. Unlike CLAUDE.md, hooks are not simply passive context that a model reads; they are an execution feature tied to a specific runtime's event model and configuration schema. The report's conclusion that Grok Build CLI does not detect these hooks is consistent with the idea that hook handling is implementation-specific rather than a shared standard. A tool can adopt the CLAUDE.md file convention while still lacking the internal event system needed to act on hook definitions.

This distinction connects to the broader context of the Model Context Protocol, or MCP, which the article frames as the common thread linking these tools. MCP is an open protocol introduced by Anthropic to standardize how AI applications connect to external tools, data sources, and services through MCP servers. It has seen growing adoption across the ecosystem, and several coding assistants now support it, which is part of why cross-tool compatibility is an increasingly practical concern. Still, MCP standardizes the interface for connecting to tools and resources; it does not necessarily standardize every convenience feature, such as context file formats or lifecycle hooks, that individual vendors layer on top of their own clients. The reported gap illustrates that shared support for MCP does not guarantee that all higher-level features transfer cleanly between products.

The practical takeaway for developers is to treat portable configuration and runtime-specific automation as separate categories. Guidance that lives in CLAUDE.md is likely to travel well, but any workflow that depends on hooks to enforce behavior, such as automatically running tests, blocking risky commands, or applying formatting, should not be assumed to work outside the tool that defines it. Teams that rely on such automation may need to reimplement it using whatever mechanism the target CLI provides, or to keep critical safeguards in tool-agnostic places like CI pipelines and pre-commit hooks in version control, rather than depending on an assistant to enforce them.

It is worth noting the limits of a single test. The findings reflect the behavior observed at the time of writing, and CLI tools in this space are updated frequently, so support for hooks or other features could change in later releases. Readers should also distinguish Grok Build CLI, xAI's agentic command-line tool, from the underlying Grok models, since the compatibility behavior described here is a property of the CLI client rather than the model itself. As the number of terminal-based AI coding agents grows, including offerings tied to Claude, Grok, and other providers, this kind of granular compatibility testing is becoming a useful way to map the real boundaries between tools. The overall lesson is less about any one product's shortcomings and more about setting accurate expectations: file-based context tends to be portable, while execution-level features remain tightly coupled to the runtime that implements them.