Repeatability Beats Folklore
Onboarding and upgrades are where agent setups die. Scripts you can dry-run, profiles, and predictable repo hygiene beat heroics and one-off tribal knowledge.

Repeatability Beats Folklore
Hero setups don’t scale. If only one person knows how the folders were created, which rules were copied, and which symlinks matter, the system isn’t portable—it’s folklore. The next hire, the next machine, the next quarter breaks what wasn’t documented as a repeatable path.
What helps: previewable installs (--dry-run), profiles (minimal vs full), predictable ignore rules for generated or local-only paths, and upgrade steps that don’t require archaeology.
Onboarding and upgrades are where agent-assisted workflows usually die—not because the model is weak, but because the setup is unique to one brain.
The insight: repeatability is a product requirement for AI-assisted engineering, not a nice-to-have. If it can’t be run twice, it isn’t a system.