Lesson 3.2 — Piping Shell Data into Claude: git diffs, logs, test output #claude #ai #piping #shell

The typical path a Claude.ai power-user takes when trying to automate diff review:
open a terminal, run git diff HEAD~1, select all 400 lines, switch to the browser,
paste into chat, read the response, switch back to terminal. Repeat for each commit.

For a 20-commit sprint review that is 40 context switches and 20 manual pastes.
It produces no persistent output — the review lives in a browser tab, not a file.
It cannot run in CI. If the diff spans multiple files and exceeds the chat input
box's paste limit, the paste silently truncates. And if a teammate needs to reproduce
the same review tomorrow, they start over from scratch.

The assumed upgrade — writing a loop in shell that calls the Claude.ai web API — hits
a wall immediately: there is no public chat API, only the Messages API, which requires
an API key and bills per token. The developer ends up at claude --print without
understanding what the flag actually controls, and the bugs begin.

Видео Lesson 3.2 — Piping Shell Data into Claude: git diffs, logs, test output #claude #ai #piping #shell канала SystemDesign Demo 1