DEV Community

Cover image for I shipped a 2-line Claude Code plugin that makes it shut up
Rodrigo Oler
Rodrigo Oler

Posted on • Originally published at github.com

I shipped a 2-line Claude Code plugin that makes it shut up

#claude #ai #opensource #productivity

Toggle it on and Claude Code answers in ≤ 5 lines — until you turn it off. Open source, MIT.

Claude Code is brilliant. It's also a chatterbox. You ask a quick thing and get three paragraphs, a recap, and a "great question!" you didn't need. Reading long terminal output, dozens of times a day, is genuinely tiring.

So I built tldr — a tiny plugin that adds a persistent ON/OFF brevity mode.

The problem

Verbosity is the default for a reason: when you don't know what the user needs, more context is safer. But once you know exactly what you want, that safety net becomes noise. And nagging "be brief" every single turn doesn't stick — the model drifts back to long-form a few messages later.

I wanted a switch, not a plea.

Install (2 lines)

In any Claude Code session:

/plugin marketplace add rodrigooler/claude-code-tldr
/plugin install tldr@claude-code-tldr
Enter fullscreen mode Exit fullscreen mode

Then just say /tldr (or "short mode", "be brief"). Say /tldr off to go back.

Before / after

Before (default):

Great question! Let me walk you through this. There are a few things to
consider here. First, the cache is stored in ~/.cache, and the reason it
grows is because... (12 more lines) ... so in summary, run `cache clear`.
Enter fullscreen mode Exit fullscreen mode

After (tldr ON):

Run: cache clear
Cache lives in ~/.cache; it grows because nothing prunes it.
Enter fullscreen mode Exit fullscreen mode

Same answer. A tenth of the reading.

How it works

This is the part I like: there's no MCP server, no background process, no telemetry. tldr is a single Skill — one SKILL.md of behavioral rules that Claude loads when you flip the mode. The whole "product" is a prompt.

While ON it enforces: max 3–5 lines, result + next step only, no filler, short bullets. And it has guardrails — code you'll paste/run and security warnings are never truncated, only the chatter around them is. For a real decision it gives the verdict in ≤5 lines and offers expand for the full detail.

The persistence is the trick: it's a mode, not a one-off. It stays on across turns until you turn it off.

What I learned shipping it as a plugin

  1. The smallest useful thing is allowed to be tiny. One Markdown file is a legitimate product if it removes real friction.
  2. Distribution is a plugin manifest. A valid marketplace.json + plugin.json turns a personal skill into a 2-line install for anyone.
  3. Validate the real flow, not the files. I installed it from GitHub into a throwaway config (CLAUDE_CONFIG_DIR) before announcing — manifests passing isn't the same as the install working.

Try it

It's free and MIT. If it saves you a scroll, a ⭐ helps me make more:
https://github.com/rodrigooler/claude-code-tldr

What Claude Code skill should I build next? Tell me in the comments.

Top comments (0)