GitHub - adamkdean/claude-chalk: A Claude Code mod that lets Claude colour its own replies in the terminal

GitHub

4 min read Original article ↗

claude-chalk: colour for Claude Code's replies, in chalk-template syntax

Licence: MIT Claude Code mod Version 0.1.0 TypeScript, strict No dependencies

Install · Examples · Palette · Syntax · How it works

claude-chalk is a Claude Code mod that lets Claude colour its own replies in the terminal. Claude writes markup in the syntax of the chalk-template npm package, and the mod draws it in colour, without the braces.

Long replies can bury the one line that matters. With the mod loaded, Claude marks failures in red, passes in green and warnings in yellow, and it puts a question it needs answered on a coloured background. The state of things shows at a glance, and most text stays plain.

Claude checks a release: three passing items with green ticks, a failed security scan in red, two open items in yellow, and a question on a blue background

Install

At the prompt of a Claude Code session in a terminal:

/plugin install chalk --marketplace adamkdean/claude-chalk

Answer y to add the marketplace, then choose a scope. The mod is active in that session at once and, under the user scope, in every session after it.

From a shell:

claude plugin install chalk --marketplace adamkdean/claude-chalk

In the desktop app, where /plugin does not run: click + next to the prompt box, then Plugins, Add marketplace with adamkdean/claude-chalk, then Add plugin.

The plugin is named chalk, after the npm package whose syntax it uses. Claude Code reserves plugin names that start with claude-.

Examples

The replies in these screenshots are scripted to show what the mod can do. Each one is played back through a live Claude Code session with the mod loaded, so every image shows exactly how the mod draws it.

Test runs

Failures stand out from passes, and the coverage gap is easy to spot.

A test run summary: 3 failures in red, 214 passed in green, 6 skipped and the coverage gap in yellow

Code review

Each finding carries its severity on a coloured background.

A code review of an average function: findings labelled HIGH on red, MED on yellow and LOW on blue, then a fixed version

Incidents

Service status reads at a glance, and the root cause is in red ahead of the detail.

Health check triage: green, red and yellow status dots per service, with the cache outage in red as the root cause

Questions that need an answer

When Claude cannot go on without a decision, the question goes on its own line on a background colour: blue for an ordinary question, red when the answer allows something risky or hard to undo.

Advice before dropping a production table, ending in a question on a red background

The palette

Every style the mod draws.

The full palette: text colours, bright colours, backgrounds, modifiers, exact hex and RGB colours, and gradients

Syntax

A block is {style text}: an opening brace, a style, one space, the text and a closing brace. Chain styles with dots. Blocks nest, and an inner block adds to the styles around it.

Kind Styles
Colours black red green yellow blue magenta cyan white gray, and each of the first eight with Bright (redBright)
Backgrounds bg and a colour: bgRed, bgBlueBright
Modifiers bold dim italic underline inverse strikethrough
Hex #ff8800 (text), #:003366 (background), #ffffff:cc0000 (both), hex(#ff8800), bgHex(#003366)
RGB rgb(255,136,0), bgRgb(0,51,102), with no spaces 
{green ✓ All 42 tests pass}
{red.bold Build failed} in `src/parser.ts`
{yellow Warning:} the cache is stale.
{bold Hello {red world}}
{#ff8800:001122.bold Hot}
{bgRed.whiteBright  Delete these 3 files? }

Inside a block, write \{ and \} for literal braces. Text in braces that is not a valid style, such as {name}, is left as written.

How it works

The mod has two hooks:

  • prompt.compose adds a short section to the system prompt that teaches Claude the syntax and when to use it. Headless runs (claude -p) do not get it, because nothing draws the colours there.
  • ui.render on AssistantMessage redraws each reply block that holds markup. A paragraph with markup is drawn as coloured text, and every other paragraph is drawn as markdown, as before. A block without markup is left to Claude Code.

Named colours are drawn as Claude Code draws those names. Hex and RGB colours are drawn exactly. Only the drawing changes: the transcript keeps the markup as Claude wrote it.

Limits

  • Markup is drawn in prose only. Inside code spans and code blocks it is shown as written.
  • A block ends at the end of its paragraph.
  • In a paragraph with markup, bold, italic, inline code, headings and lists are drawn, but tables are shown as plain text.
  • Where the mod is not loaded, the markup is shown as written.
  • The mod API is in early access and can change between Claude Code releases. Tested with Claude Code 2.1.312.

Development

claude --plugin-dir .      # start a session with the mod loaded from this folder
claude plugin validate .   # check the manifests and the hooks module
claude plugin test .       # run the tests in tests/

AGENTS.md describes the layout, the checks and the rules for changes, including how to check drawing in a real terminal.

The screenshots play scripted replies through a live session in a pseudo-terminal, record the output, and replay it into xterm.js to render it.

Legal

Released under the MIT licence.

An independent project, not made by or affiliated with Anthropic or the chalk project.