harness: Skill for Claude Code

Analyze Image Size

Produces a three-part breakdown of the harness Docker image:

1. All available tags and their sizes
2. Per-layer contribution (from docker history)
3. Efficiency and waste report (from dive --ci)

Step 1: Determine target image

Default to ghcr.io/boldblackai/harness:latest unless the user specifies a different tag.

Step 2: List all harness image tags

docker images ghcr.io/boldblackai/harness --format "table {{.Tag}}\t{{.Size}}\t{{.CreatedAt}}"

Step 3: Per-layer breakdown

docker history <image> --no-trunc --format "table {{.Size}}\t{{.CreatedBy}}"

Most layers will be 0B (metadata). Focus the summary on layers with non-zero size. Translate the raw CREATED BY commands into human-readable descriptions:

Step 4: Base image size

Read Dockerfile to find the FROM line and extract the base image reference (including the pinned digest). Pull it if needed and inspect its uncompressed size:

docker pull <base-image-with-digest> 2>/dev/null
docker inspect <base-image-with-digest> --format '{{.Size}}' | awk '{printf "%.0f MB\n", $1/1024/1024}'

Step 5: Dive efficiency analysis

docker run --rm \
  -v /var/run/docker.sock:/var/run/docker.sock \
  wagoodman/dive:latest --ci <image>

This outputs:

• Overall efficiency percentage
• Total wasted bytes
• A ranked list of inefficient files (files duplicated across layers)

Step 6: Report

Present the results as a unified report:

## Image size analysis: <image>

### Tags
<table of tag / size / created>

### Layer breakdown
| Size   | Layer                        |
|--------|------------------------------|
| 100 MB | Base OS (debian:stable-slim) |
| 249 MB | System packages + Node.js    |
| 429 MB | pnpm + agent packages        |
| ~1 MB  | Config files / entrypoint    |

Base image accounts for X MB of the Y MB total.

### Efficiency (dive)
- Efficiency: XX%
- Wasted: XX MB

Top waste sources:
| Wasted | File | Likely cause |
|--------|------|--------------|
...