The AI-written shell script that ran from the wrong folder and exited 0: `set -euo pipefail`, and where it still won't save you

A developer demonstrated that a four-line bash script with an unset BUILD_DIR variable continued past a failed cd, ran its deploy line, and exited 0 until strict mode was added. Prepending set -euo pipefail made the script halt on the first error with exit code 1 and an unbound variable message. The writeup also notes that set -e does not trigger on failures inside if conditions, while/until tests, non-final commands in && or || lists, or negated commands, and recommends agents generate scripts with strict mode and quoted variables while humans review cd, rm and deploy lines.

Result first: a four-line script whose cd failed because BUILD DIR was unset kept going. It printed the wrong working directory, ran its "deploy" line, and exited 0. Adding one line, set -euo pipefail , made it stop on the first error with exit code 1: BUILD DIR: unbound variable . bash /usr/bin/env bash set -euo pipefail cd "$BUILD DIR/out" What each flag does bash's own help set : -e : "Exit immediately if a command exits with a non-zero status." -u : "Treat unset variables as an error when substituting." -o pipefail : "the return value of a pipeline is the status of the last command to exit with a non-zero status, or zero if no command exited with a non-zero status." Without it, false | true counts as success. Where -e still won't save you GNU Bash manual : the shell does not exit when the failing command is part of the test in an if , inside a while / until condition, any command in a && or || list except the final one, or a command inverted with . The demo shows if false; then …; fi carrying on under set -e . Don't treat strict mode as a substitute for checking the commands that matter. For AI-assisted scripting: ask your agent to start every bash script with set -euo pipefail , and to quote variables. Then read any cd , rm or deploy line yourself before running it. Do this now: run head -3 on your repo's .sh files and add strict mode to any that lack it. Then test each one, because scripts that relied on ignored errors will now stop. From Penloom Studio's free AI lessons: penloomstudio.com https://penloomstudio.com/landing-cogloom-motion.html