A Claude Code hook to rewrite the model's repeated verbal tics

Sources: write-up, HN 48905248
Summary: A post by Johanna Larsson reached the Hacker News front page describing recurring stock phrases in Claude output such as "load-bearing" and "you're absolutely right" and a client-side fix. The fix is a Claude Code display hook, a Python script placed in ~/.claude/hooks/ and registered in settings that intercepts output and substitutes chosen words before display, without changing the model.
Comments: The thread treats the verbal tics as a widely shared annoyance and debates whether hook-based text substitution is a real fix or a cosmetic one, since it edits the presented text rather than the generation.
Why it matters: It is a small concrete example of using Claude Code hooks to shape agent output locally, and it surfaces how uniform model phrasing has become across users.