MCP: drift_images returns empty results for harbor stacks

[gfargo]

Summary

strut_drift_images returns total: 0, drifted: 0, images: [] for stacks on harbor that have many running containers with real images. Image introspection appears to not work for harbor-hosted stacks.

Steps to Reproduce

1. Call strut_drift_images with stack: "agent-platform" (8+ containers on harbor)
2. Observe: returns {"total": 0, "drifted": 0, "unknown": 0, "images": []}
3. Call strut_drift_images with stack: "immich" (4 containers on harbor)
4. Observe: same empty result

Comparison

strut_drift_images on octoprint-ender3 (watch) correctly returns 1 image (status: unknown). So the tool works on watch but not on harbor.

Expected Behavior

Should enumerate all images used by the stack's running containers and check if newer digests are available upstream.

For agent-platform, should report images for: supervisor, webapp, electric, postgres, redis, clickhouse, minio, docker-proxy (at least 8 images).

Possible Causes

• SSH to harbor succeeds (briefing/health work) but the image introspection script may fail silently
• Different Docker/Compose version behavior on harbor vs watch
• The compose project name mismatch (same root cause as MCP: health check false negative - 'No containers running' despite active services #502) preventing container enumeration

Environment

• strut version: 0.45.4, MCP server via strut mcp serve
• harbor: Lenovo M70q, 16GB, Docker v29.4.1
• watch: Pi 4B, 4GB, Docker Compose v5.1.3

[gfargo]

Resolved by #519, which adds fallback image-container enumeration when project-name matching returns no rows. Closing as fixed.