code-provenance

Resolve Docker images in a docker-compose file to their exact source code commits on GitHub.

Available as both a Python package and a Node.js package.

Installation

Node.js:

npm install -g code-provenance

Python:

pip install code-provenance

Quick Example

Point it at a SecretVM machine and it fetches the running docker-compose and resolves every image:

code-provenance secretai-jedi.scrtlabs.com

Or resolve a local docker-compose.yml:

version: "3"
services:
  web:
    image: traefik:v3.6.0

code-provenance --workload docker-compose.yml

Output:

web: traefik:v3.6.0
  repo:       github.com/traefik/traefik
  commit:     06db5168c0d9
  status:     resolved
  confidence: exact
  url:        https://github.com/traefik/traefik/commit/06db5168c0d9...

How It Works

The tool resolves each Docker image through a chain of strategies, stopping at the first success:

1. OCI labels -- reads org.opencontainers.image.source and org.opencontainers.image.revision from the image manifest
2. Repo inference -- maps the image registry/namespace to a GitHub repository (GHCR namespaces map directly; Docker Hub images are resolved via the Hub API)
3. Tag matching -- matches the Docker tag against git tags in the inferred repository
4. Packages API -- for GHCR images, queries the GitHub Packages API to resolve digests or :latest tags
5. Latest release / commit -- for :latest tags, falls back to the most recent GitHub release or the HEAD commit on the default branch

Confidence Levels

• exact -- the commit is definitively tied to the image (e.g., OCI revision label with a versioned tag, or an exact git tag match)
• approximate -- the commit is a best-effort match (e.g., :latest tag, prefix tag match, or latest release fallback)

CLI Usage

code-provenance [machine-url] [--workload FILE] [--image IMAGE] [--json] [--verbose]

Argument	Description
machine-url	SecretVM machine URL or hostname (e.g. secretai-jedi.scrtlabs.com). The docker-compose is fetched from it over port 29343, falling back to 21434 for GPU machines.
--workload FILE	Resolve a local docker-compose file instead of a machine URL
--image IMAGE	Resolve a single image reference
--secretvm URL	SecretVM machine URL (same as passing it positionally; kept for backward compatibility)
--json	Output results as JSON
--verbose, -v	Show the resolution steps for each image

Examples

# Resolve all images running on a SecretVM machine (default mode)
code-provenance secretai-jedi.scrtlabs.com

# Resolve all images in a local compose file
code-provenance --workload docker-compose.yml

# Resolve a single image
code-provenance --image ollama/ollama:0.12.3

# JSON output
code-provenance secretai-jedi.scrtlabs.com --json

# Show resolution steps
code-provenance --workload docker-compose.yml --verbose

Authentication

The tool works without authentication for basic resolution (OCI labels, tag matching). However, a GITHUB_TOKEN is required for:

• Resolving digest-only images (e.g., image@sha256:...) via the GitHub Packages API
• Resolving :latest tags on GHCR images via the Packages API
• Higher GitHub API rate limits (5000/hr vs 60/hr unauthenticated)

Setup

Option 1 — Use the gh CLI (recommended): If you have the GitHub CLI installed and authenticated, run.sh auto-detects the token. Make sure your token has read:packages scope:

gh auth refresh -h github.com -s read:packages

Option 2 — Set the environment variable directly:

export GITHUB_TOKEN=ghp_your_token_here

Create a classic token at https://github.com/settings/tokens with read:packages scope.

Library Usage

Both packages export the same API for programmatic use.

Node.js:

import { parseCompose, parseImageRef, resolveImage } from "code-provenance";
import { readFileSync } from "fs";

const yaml = readFileSync("docker-compose.yml", "utf-8");
for (const [service, image] of parseCompose(yaml)) {
  const ref = parseImageRef(image);
  const result = await resolveImage(service, ref);
  console.log(`${result.service}: ${result.commit} (${result.confidence})`);
}

Python:

from code_provenance.compose_parser import parse_compose, parse_image_ref
from code_provenance.resolver import resolve_image

yaml_content = open("docker-compose.yml").read()
for service, image in parse_compose(yaml_content):
    ref = parse_image_ref(image)
    result = resolve_image(service, ref)
    print(f"{result.service}: {result.commit} ({result.confidence})")

See python/README.md and node/README.md for full API reference.

License

MIT