hush
You asked for a change. Read what changed.
Available on
 |
 Claude Code |
Hush has no Codex package.
Get started · What is this? · How it works · What you can do · Evidence
What is this?
You ask the assistant to fix a failing test. It reads files, runs commands and explains each step. By the time it finishes, the result is buried in the conversation. Hush reduces that reading.
It quiets routine narration, shortens noisy tool output and shapes the final answer around what changed, whether it worked and what comes next. The example illustrates the workflow; the measurements below describe recorded sessions.
Why you'd want it
- Read the result with less play-by-play.
- Keep noisy command output from crowding the conversation.
- Inspect retained full output when you need more detail.
How it works
A writing style asks the assistant to stay quiet during routine work and finish with a short, useful answer. Session controls reinforce that behavior and trim selected tool results. Large outputs can be stored temporarily, with a reference for closer inspection.
What you can do
| You want to | Outcome |
|---|---|
| Read the result first | Short, plain answers focused on the result |
| Inspect shortened output | A reference to the retained full result where available |
| Turn off a session control | One environment variable per control, listed in Settings |
Install
Hush runs in Claude Code. It is not available for Codex.
Claude Code
Inside Claude Code:
/plugin marketplace add V-Songbird/foundry
/plugin install hush@foundry
Start a new session to load the plugin. Hush's voice applies on its own. The published numbers also
set "outputStyle": "hush:Hush" in ~/.claude/settings.json, which adds Claude Code's per-turn
reminder that the style is active; see Settings.
Requirements: Claude Code 2.1.139 or later, and Node.js 22 or later.
Good to know
Correctness comes before silence. A short answer can omit a useful detail, and quieter sessions do not always cost less. Ask for depth when you need it, and check the result before acting.
Full-output references point to temporary files. Disabling the session controls leaves the writing style on; disable the plugin to turn it off. See Settings for switches, and How hush works for how long those files are kept and who can read them on each platform.
The numbers
The comparison asks whether jobs were completed correctly, how often the assistant spoke at most once before the answer, how long its final answers were and what a session cost. These observations do not guarantee the same behavior in your sessions.
Claude Code results
These results measure the voice released in 1.13.0, not the current one. The current voice also caps each block of the final answer at 40 words, and these figures do not measure that rule. A block is the text between two blank lines. The rule was tested in a separate batch; see capping blocks at 40 words.
| Model | Setup | Jobs right | Spoke at most once before the answer | Median final prose words | Average bill per session |
|---|---|---|---|---|---|
| Claude Opus 5.5 | No plugin | 36/36 | 28/36 | 369 | $0.421 |
| Claude Opus 5.5 | hush | 36/36 | 36/36 | 271 | $0.458 |
Watch the recorded Claude Code demo: one real session each on Claude Opus 5.5, paced by each session's recorded wall-clock time
Both setups passed all 36 task checks. Hush spoke at most once before the answer in every session, and said nothing at all before it in 34 of 36. Without a plugin, Opus 5.5 already spoke at most once in 28 of 36 sessions.
Hush's median final prose was 271 words against 369 without a plugin, about 27% shorter. Its sentences were shorter too: 0.5% ran past 20 words, against 10.9% without a plugin. Runnable content appeared in all 36 answers of both setups; that text heuristic does not check that a command is correct or complete.
On Opus 5.5, hush costs more. The average bill per session was 9% higher. Seven of the nine jobs cost more, by 4% to 32%, and four of those are above about 16%, the largest gap that batch showed between two near-identical setups. Only the two jobs whose commands print the most came out cheaper, by 19% and 4%. Opus 5.5 already writes less without a plugin, so hush has less to cut. See cost and runnable content on Opus 5.5.
Batch o55-451-495ce18f: nine fixture jobs, four runs per setup, in Claude Code with Claude Opus 5.5 (model id claude-opus-5-5) at the default effort, September 28, 2026. Measurements used HUSH_WRAP=1. Word counts exclude fenced code. These measurements describe the recorded Claude sessions, not Codex performance. The earlier voices, measured on Opus 5 and Sonnet, are in the benchmark details; the records and definitions behind them are kept outside this repository.
Results can vary between runs.
Going deeper
How it works · Settings · Benchmark details · Changelog
Foundry lists this plugin. The research records and the benchmark harness are kept outside this repository and are not distributed with the plugin.
License
MIT — see LICENSE.