fix: publish-hostless workflow picks wrong container image from ECS

[field123]

Problem

The "Publish Hostless Packages" GitHub Actions workflow fails with exit code 127: node: command not found.

This has been broken since March 2026. Last successful run was February 19, 2026.

Root Cause

The workflow queries the deployed ECS task definition to find the WAB container image:

--query 'taskDefinition.containerDefinitions[0].image'

But containerDefinitions[0] is the Fluent Bit log-router sidecar (amazon/aws-for-fluent-bit:stable), not the WAB container. The task definition has 3 containers:

Index	Name	Image
0	log-router	amazon/aws-for-fluent-bit:stable
1	wab	941131938315.dkr.ecr.../plasmic/wab:...
2	datadog-agent	public.ecr.aws/datadog/agent:latest

The Dockerfile.publish-hostless then builds on top of the Fluent Bit image (which has no node binary) instead of the WAB image, causing node: command not found.

Verified locally

# WAB image — node exists
$ docker run --rm --entrypoint /bin/sh <wab-image> -c "which node"
/usr/local/bin/node

# publish-hostless image (built on Fluent Bit) — no node
$ docker run --rm --entrypoint /bin/sh <publish-hostless-image> -c "which node"
(nothing found)

Fix

Query by container name instead of array index:

--query "taskDefinition.containerDefinitions[?name=='wab'].image | [0]"

Why it worked in February

Before Datadog sidecar containers were added to the WAB task definition, containerDefinitions[0] was the WAB container. When sidecars (Fluent Bit, Datadog) were added, WAB shifted to index 1 but this query was not updated.