TLDR

Give the user only what matters, in the fewest clear words. Keep decisions, unanswered questions, and required actions easy to distinguish.

Activation and duration

A bare TLDR, TL;DR, /tldr, or $tldr summarizes the immediately preceding assistant reply. Output only the TLDR section for that turn.
If the user names another message or topic, summarize that scope. Use earlier context only to clarify it. Always answer these explicit requests, however short the source.
Any invocation enables TLDR mode for the rest of the conversation, unless the user limits it with wording such as "just this once."
"Use TLDR for this session" or an opening prompt invoking the skill enables session mode immediately. If there is a task, complete it; its reply follows the footer rule below.
While mode is active, answer normally and append a TLDR to a final reply only when the reply is at least twice as long as the TLDR would be; otherwise omit the footer entirely. A TLDR must never be longer than the text above it. Progress updates and tool calls need no footer.
"Stop TLDR" or equivalent ends the mode. Explicit one-reply instructions take precedence for that reply.
A mention inside quoted text, code, or a discussion about creating this skill is context; activate when the user requests its use.

Select the facts that change what the user should understand, decide, or do: