Claudette: Make Claude stop talking like a BuzzFeed article
Points and comments are a snapshot, not live.
Claudette pipes Claude's verbose output through Gemini's Antigravity CLI to strip millennial-buzzspeak prose.
The skill /debuzz (nicknamed Claudette) translates Claude's last response via `agy -p "..."` into plain English in three modes: colleague, manager, or director. The author explains that no amount of prompting cures Claude's TED-talk tone, so the honest fix is to hand the response to a different model and print its translation verbatim. A before/after example shows the same bug report transformed from a numbered list of 'load-bearing assumptions' into a direct file-path, error-type, and fix.
The skill runs as a Claude Code skill; it writes the previous reply to a temp file, pipes it through the Antigravity CLI, and prints the output verbatim. If agy errors, Claude's own rewrite is shown as a clearly labeled fallback.
What commenters are saying
Many commenters share the author's frustration, calling Claude's style tiresome, moralizing, and increasingly verbose. Some attribute it to reinforcement learning, others to Anthropic's deliberate anthropomorphization or to training data (e.g., social-media engagement patterns). Several note that switching to other models (Gemini, Codex/Sol, GLM5.3) felt like a breath of fresh air. A few offer workarounds: pruning Claude.md, using system prompts to forbid certain phrasing, or employing separate tools like 'bro' or 'Claudish to English'. One commenter argues the verbosity helps Anthropic charge more tokens; another suggests it serves as unintended watermarking.