A Claude Code skill rewrites Claude's verbose replies via Gemini CLI
The /debuzz command hands Claude's last message to Gemini for a plain-English rewrite — three audience modes, a one-line install, and a real privacy trade-off.
Copy markdown
One command reroutes Claude's last reply through Gemini
`/debuzz` writes Claude's previous message to a temp file, runs it through the Gemini CLI with plain-English instructions, and prints Gemini's version verbatim — Claude's own rewrite only appears as a clearly labeled fallback if Gemini errors. It targets the theatrical, numbered-revelation style Claude Code users keep complaining about.
Three modes tune the rewrite to your reader
`/debuzz [mode] [text]` takes colleague (the default, keeps technical detail), manager (high-level overview), or director (executive summary). Run it with no text and it rewrites Claude's last message; pass your own text and it will debuzz anything you hand it.
The catch: your text goes to Google
It shells out to an authenticated gemini-cli (`npm install -g @google/gemini-cli`), so anything you debuzz is sent to Gemini — worth knowing before piping sensitive output through it. Install is a git clone into `~/.claude/skills/`, and it's free if you already hold a Gemini key.
Elsewhere: Claude Code plugs a long-session memory leak
Claude Code 2.1.238 fixed unbounded memory growth in long interactive sessions — subagent tool results are now freed once they scroll out of the recent display window. The same build adds a `keybindingFlavor: "readline"` option so Ctrl+W deletes back to the previous whitespace, Bash-style.