For agent-led teams Why start from a built image
Agents are fast at git clone. What they hand you a few minutes later is a build nobody else
has seen: produced on one machine, from whatever the registries served that minute, with choices nobody
wrote down. An image settles the six steps between source and running once, in the open.
A starting state, not a build
pull by digest is one step that either matches or doesn't.
Building from source is dozens of steps that can each go differently on a different day,
with a different toolchain, a different lockfile resolution, a different maintainer upstream.
Trust you can check
Signatures verified against a declared identity, SBOMs you can download, build provenance where the publisher ships it,
and a dated grade whose arithmetic is on the page. Evidence your security team can inspect, not a vibe.
An open rubric, dated
Every grade is a formula with the ingredients listed: CVE count by severity, how fresh the image is,
whether it declares a healthcheck, whether its signature verifies.
The methodology page has the full changelog.