Skip to content

Stage 2 ran against a real daemon, and both live tests pass - #240

Open
thedancingdeveloper wants to merge 1 commit into
mainfrom
docs/stage-2-live-evidence
Open

Stage 2 ran against a real daemon, and both live tests pass#240
thedancingdeveloper wants to merge 1 commit into
mainfrom
docs/stage-2-live-evidence

Conversation

@thedancingdeveloper

Copy link
Copy Markdown
Contributor

The first execution of any part of this repository against a real Docker daemon — on Node B, in the deployed stack, against a dedicated DinD daemon beside the controller.

success: True
..                                                            [100%]

An agent reads and writes its own worktree, reads a declared mount, cannot reach an undeclared sibling path that exists on the host, receives none of the controller's credentials while still receiving explicitly passed variables, is denied the network under network=none and allowed it under bridge, and its container is gone after teardown.

The evidence package records image digests, the Komodo update id and the reproduction command — "it passed" is not reproducible and a tag moves.

The Stage 2 exit is NOT claimed

  • The security profile itself (no-new-privileges, dropped capabilities, read-only rootfs, resource limits, recorded digest) is asserted only in the docker create argv and never verified from inside a live container.
  • The nested daemon's own confinement is untested.
  • One sandbox image, Alpine/BusyBox. The last defect this suite found was an assumption about exactly which userland is present.
  • No workload ran; the deployed controller has no model routes and cannot claim work.

Also here

The build now pins and bumps the acceptance image by sha. It pinned :latest, and docker compose run does not re-pull a tag it already holds — so the live suite silently re-ran a stale image and reproduced a defect that had already been fixed and deployed (:latest was sha256:d73e54bd…, the current build sha256:9808728b…). An acceptance run that cannot name the image it ran is not evidence.

The first execution of any part of this repository against a real Docker
daemon. On Node B, in the deployed stack, against a dedicated Docker-in-Docker
daemon beside the controller:

    success: True
    ..                                                            [100%]

An agent reads and writes its own worktree, reads a declared mount, cannot
reach an undeclared sibling path that exists on the host, receives none of the
controller's credentials while still receiving explicitly passed variables, is
denied the network under `network=none` and allowed it under `bridge`, and its
container is gone after teardown.

The evidence package records the image digests, the Komodo update id and the
command to run it again, because "it passed" is not reproducible and a tag
moves.

**The Stage 2 exit is not claimed.** The security profile itself —
no-new-privileges, dropped capabilities, read-only rootfs, resource limits —
is asserted only in the `docker create` argv and never verified from inside a
live container. The nested daemon's own confinement is untested. One sandbox
image has been used, and the last defect this suite found was an assumption
about exactly that. No workload ran.

Also here: the build now pins and bumps the acceptance image by sha. It pinned
`:latest`, and `docker compose run` does not re-pull a tag it already holds --
so the live suite silently re-ran a stale image and reproduced a defect that
had already been fixed and deployed. An acceptance run that cannot name the
image it ran is not evidence.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant