A strange behavioral quirk in Claude Code is making the rounds: the tool apparently refuses certain requests, or triggers a different billing response, when your git commit messages contain the string "OpenClaw."
Developers noticed the pattern when working on projects with that keyword in their commit history. Claude Code, Anthropic's terminal-based coding assistant, reads surrounding context from your repository when responding to requests - including, it appears, commit messages. Something about the "OpenClaw" string is triggering a guardrail or a pricing branch that shouldn't activate under normal use.
What's unclear is whether this is intentional. If Anthropic deliberately built in behavior that scans commit history for specific keywords and modifies responses accordingly, that's a product policy worth explaining publicly. If it's a bug in how the context window (the amount of text the model reads before responding) is being parsed, it's a significant one - users have no reason to expect their commit message wording to affect whether a tool works or what it costs.
Anthropics hasn't issued a public statement on the report. For now, developers hitting this issue have a simple workaround: rename the reference in commits or avoid the keyword while the behavior is unresolved. That's not a satisfying answer, but it's the practical one until there's an official explanation.
This also raises a broader question about AI coding tools that read repository context: how much of your project's history and metadata is being processed, and what else might be in there that changes tool behavior in ways you'd never anticipate?