# terse

> A Claude Code plugin that shortens replies and removes mannered prose such as slogans, metaphors and "X, not Y" framing.

terse is a Claude Code plugin by lowenbjer that applies a set of writing rules to Claude's replies, documents, commits and subagent output. The README reports that it cuts reply length roughly in half and ships with an optional context meter for the status line. It is released under the MIT license; the latest release listed is v1.2.2.

## What It Is

terse is a writing-style plugin for Claude Code. It targets verbose, essay-like answers and mannered prose: slogans, metaphors, recaps of what was just done, closing offers and "it's not X, it's Y" framing. The goal is a reply that gets to the point on every turn.

## How the rules are applied

The rules ship as an output style that becomes part of the main agent's system prompt. The README describes the reply shape as one sentence with the answer, then a list or table, then at most one caveat sentence. It lists 22 rules and 18 before/after pairs, with chat replies capped at 150 words unless a document or walkthrough is requested.

Subagents run their own system prompt, so a SubagentStart hook passes them the same text. A UserPromptSubmit hook adds a condensed reminder to each prompt to counter style drift over long sessions, at about 145 input tokens per turn according to the README. The plugin makes no network calls and collects no data.

## Context meter

Running /terse:install-meter copies a status line script and adds a statusLine entry to the user's Claude settings. It shows the model name and context fill as a percentage with color thresholds at 37%, 49% and 60%.

## Measured results and limits

The author reports benchmark figures: on a 20-prompt run against a real codebase, 46% fewer words on Fable 5.1 and 54% fewer on Opus 5.5. On 12 public prompts with Opus 5.5, the README reports a 70% word reduction for chat replies. The README also documents limits: the 150-word cap shortens replies without strictly enforcing it, subagent output follows the subagent model, and cost falls less than word count.

## Setup path

Install from the Claude plugin directory listing, or from GitHub with /plugin marketplace add lowenbjer/claude-terse followed by /plugin install terse@terse. Uninstall is done through the plugin settings or /plugin uninstall.

## Features
- Output style with 22 writing rules for shorter, plainer replies
- Reply shape: answer sentence, list or table, at most one caveat
- SubagentStart hook applies the rules to subagents
- UserPromptSubmit hook reminder to prevent style drift
- Optional status line context meter with color thresholds
- Benchmarks and per-model tables for contributing new model results
- No network calls or data collection

## Integrations
Claude Code, Claude plugin directory

## Platforms
WEB, API, CLI

## Pricing
Open Source

## Version
v1.2.2

## Links
- Website: https://github.com/lowenbjer/claude-terse
- Documentation: https://github.com/lowenbjer/claude-terse#readme
- Repository: https://github.com/lowenbjer/claude-terse
- EveryDev.ai: https://www.everydev.ai/tools/terse