v0.22.1

One fix, for ecs-ec2 deployments that run the fleet’s own inference engines. Taking a model in failed before it fetched a byte on any deployment that never picked an engine tools image by hand, and the only line that reached the Console pointed at the documentation instead of at the cause.

Fixed

• Taking a model in failed on both roles wherever EngineToolsImageTag still held the value it came with: the template asked for a newer script contract than that image speaks, so the fetch and upload steps refused to start. 0.21.0 and 0.22.0 are both affected, and a deployment that chose its own tag never was.

Upgrade notes

• Updating is enough. update.sh moves a stack still on that first value onto the matching image and names the parameter, so it does not come back. A tag you chose yourself is left alone — rebake it from this release’s deploy/aws/ecs/engine-tools/ if you keep your own.

• Staying on 0.21.0 or 0.22.0? One parameter repairs it, and nothing else in the stack changes:

  aws cloudformation deploy --stack-name af-ecs-engines \
    --template-file deploy/aws/ecs/cfn/60-engines.yaml \
    --parameter-overrides EngineToolsImageTag=2026-09-18 \
    --capabilities CAPABILITY_NAMED_IAM --no-fail-on-empty-changeset

Agent CLI pins are unchanged from 0.22.0.