cc-stretch shows 17 percent of a Claude Code reply by default

A coding session with Claude Code often ends with a reply the length of a short blog post, and the reader has to scroll back through it to find the one line that mattered. pro-vi published cc-stretch, an MIT-licensed plugin that formats every reply as stretchtext: the reader sees a short version by default, and each marked phrase grows in place when they click it. The idea is Ted Nelson's from 1967, applied to the terminal.

How a reply is written

The plugin tells Claude to write every reply with inline markup of the shape {{short|long}}. The short half is what shows up in the skim, and the long half is kept hidden until a click. Stretches can nest, so a long paragraph can itself hide its own longer paragraphs, and a reader opens only the depth they need. A clicked phrase expands around the reader's current position so the surrounding text does not jump away. A collapse marker closes it again. While the reply is still streaming in, every stretch shows as its short form rather than flashing between states.

17 percent of the words, two of three points

The README states that cc-stretch was tested on "69 real replies from four of my coding projects." The condensed version was "17 percent of the words Claude wrote without the plugin," and that condensed version carried "about two of every three most important points." The remaining third of the important points sat inside the stretches that the reader could choose to open. The author labels the three states on the project page: skim is the view with every stretch closed, full is every stretch open, and read is what the user actually consumes, which is the skim plus whichever stretches they opened.

The project also folds completed tool calls to a single-line summary, so a reply that includes a long shell transcript or a file listing stops filling the window after the step is done.

Tokens stay the same

The README is explicit on the limit: "You read less; you don't pay less." Claude writes every level up front, so the full reply, with every stretch's long form, is sent on each turn and billed as such. The plugin saves reading time and screen room. Clicks are stated as unreliable on Windows, Claude's desktop app does not render the tool-call fold, and in a desktop session attached to a terminal the markup can show as raw text while the reply is still streaming.

Installation is a plugin command inside Claude Code, and the project is MIT licensed. The GitHub API reports 108 stars and 7 forks, and the repository is less than a day old, so the traction above is from the first day of the project being public.

A working developer who already reads long Claude Code replies on hard tasks has one new choice: let the model write the whole answer, see one-sixth of it to start with, and open only the parts worth opening. If the two-of-three figure holds on your own replies, the saved reading time is the gain. Run the plugin on your own project before deciding, since the measurement is one author's and the sample is 69 replies.

Source

• pro-vi/cc-stretch on GitHub, MIT License