G-16Environment Bootstrap

Clone, install, build, seed, green — everything that has to work before an agent can begin.

Step-by-step walkthrough7 steps

A background agent wakes up in a container with nothing in it.
Before a single line can be read, five things have to work.
Fixtures and a green baseline included — an agent that cannot run the tests has no verifier at all.
This is where autonomous runs die. Not on the task — on a native dependency that needed a system package.
The fix is to make setup a script in the repo, not lore in someone’s shell history.
Run it in CI as well, so it breaks in a pull request instead of in an agent session at 3am.
Only then does the agent start — on a baseline that is known-good rather than assumed.

Key insight

Autonomous coding runs fail here more than anywhere else, and never on the interesting part. A setup script in the repo, exercised by CI, turns the most common failure into one that breaks in a pull request instead of at 3am.

Failure mode

Setup that lives in someone’s shell history. It works on the machine it was written on and nowhere else, and the agent’s transcript is forty minutes of dependency errors.

G-16Environment Bootstrap

Elsewhere in the atlas