tldr

tldr

A toggle for people tired of bible-sized AI responses.

One slash command · ~93 tokens per message when on, zero when off · No per-project setup · Persists across sessions

You asked Claude what time it is. Three paragraphs later, you have a brief history of horology, an explainer on UTC offsets, and a caveat about leap seconds. You wanted "3:47pm".

tldr is a tiny Claude Code plugin that adds one slash command: /tldr:tldr. Flip it on and Claude defaults to 1-3 sentence replies for every subsequent message. Ask for "detail" or "explain" and you get the full essay anyway — your call, every turn.

Why install

• You ask short questions and want short answers
• You're reviewing diffs or debugging — every "let me first explain my approach…" preamble costs you 20 seconds
• You want a context-friendly default, with a one-word override when you actually do need the long version
• You don't want to retype "be brief" at the top of every message
• You'd rather burn 93 tokens reminding Claude to be terse than 900 reading the answer

Install

/plugin marketplace add https://github.com/marcelopaniza/tldr
/plugin install tldr@tldr

Use

/tldr:tldr           # toggle (default)
/tldr:tldr on
/tldr:tldr off
/tldr:tldr status

State lives at ~/.claude/plugins/data/tldr/state and persists across sessions and projects. Toggle once, applies everywhere until you flip it off.

Use cases

• Quick syntax lookups — "regex for a UUID v4" → the regex, not a history of regular expressions
• Stand-up summaries — paste a transcript, get a 2-line summary, not a 6-paragraph minutes doc
• Code review — Claude flags the issue, you read your diff, not an essay on best practices
• Incident triage — what broke and what to roll back, in one screen
• Long sessions — every reply is small, so /compact waits longer
• Quick fact checks — "does Python have ternary?" → "yes: x if cond else y"

What it costs

State	Per user message	Why
Off	0 tokens	Hook exits silently before Claude sees anything
On	~93 tokens	Hook prepends a short instruction telling Claude to be terse, with an exception clause for when you ask for detail

No persistent context, no system-prompt rewriting, no daemon — just a UserPromptSubmit hook that reads a one-byte state file.

The escape hatch

If your message contains any of these phrases, tldr yields and Claude gives you the full answer for that turn:

• "explain"
• "in detail"
• "in full"
• "walk me through"
• "deep dive"
• "step by step"

So tldr is on most of the time, and you opt back into the bible when you actually need it.

Requirements

• Claude Code with plugin support
• bash (any reasonably modern version)
• That's it.

Local development

Test without installing:

claude --plugin-dir /path/to/this/repo