Test the cd guard in a real container instead of Windows bash

main

Commit

HighnessAtharvayesterday

The old live test ran the guard through the host's bash with a Windows path and failed on quoting. Production runs steps inside the Linux container, so the live test now does the same. 205 offline and 22 live tests pass.

Entire-Checkpoint: 01M4JNS1BNG2GM1RZHKJD4BCHR

Checkpoints

Polish Website UI with Apple-Inspired Design

Claude CodeSonnet 5.5
View session
Checkpoint 1

FirstRun neatHack Kickoff and Build

Claude CodeOpus 5.5
View session
Checkpoint 1