What SCG checks—and what it does not.
SCG helps you compare recorded dependency fingerprints with upstream observations and review certain forms of credential exposure. Its usefulness depends on a baseline you have reviewed, the references it can resolve, and how you integrate the result into your workflow.
Dependency checksSecret checksCoverageData sent to the platformThe trust model
Dependency checks compare records, not intentions.
scg init creates a signed record of resolved references. scg check compares the entries in that record with the SCG platform’s available observations. A changed fingerprint is reported as drift. Incomplete verification must not be treated as a successful check.
The current check does not intercept every installation, verify every byte on disk, or monitor processes. It also does not establish that the dependencies currently declared in every project file are identical to the entries in an older baseline. Review configuration changes and regenerate the baseline deliberately when dependencies change.
A passing result is not a malware-safety verdict. A harmful package can be present when the baseline is first created, and an update can introduce harmful code even when its version and published fingerprint are internally consistent. Keep vulnerability scanning, package-manager integrity checks, immutable pins, and review in place.
Secret checks support a permissions review.
The CLI’s workflow secret checks compare credential-like environment-variable names with patterns in a tool’s security profile. The result depends on which profile is available and which environment variables the SCG process can see.
This is not a sandbox or a general guarantee that a tool cannot read a file, a socket, process memory, or another credential source. Do not assume that running SCG in one shell command removes credentials from a later command or a separate workflow step. Restrict credentials in the actual execution environment.
A local audit cannot see credentials that exist only in a remote build environment. A profile derived from a tool’s manifest is also not the same as an independently reviewed least-privilege policy.
Check the files and references your release supports.
The reviewed CLI supports GitHub Actions, npm registry dependencies, Python package references, and supported Dockerfile image references. Input parsers include npm package lockfiles and pnpm lockfiles. Support for an ecosystem does not mean that every syntax, registry, package variant, or reference is covered.
- GitHub Actions: YAML workflow files under
.github/workflows. - JavaScript:
package-lock.jsonand supportedpnpm-lock.yamlformats. - Python:
requirements.txtandrequirements-*.txt; prefer exact versions. Requirement ranges are not a substitute for a fully resolved dependency lockfile. - Containers: supported Dockerfile
FROMreferences, within the platform’s available registry coverage.
Native Windows releases, Go-module checks, Helm checks, arbitrary installed-application inventory, and universal private-registry support are not promised here. Follow the released CLI documentation rather than assuming that platform indexing implies end-to-end CLI support.
Initialization reads the supported project files in the project root, with or without a .github/workflows directory. A project with no supported dependency is told so rather than given an empty baseline, and a requirement the parser cannot reduce to one artifact, such as a Python version range, is reported as not recorded instead of being recorded as if it were pinned.
Local execution still uses a hosted service.
Dependency resolution sends references to the platform. Baseline creation sends the lockfile content for signing. Registering a watched repository uploads the signed lockfile and associates it with your organization.
Lockfile metadata can contain dependency names and versions, fingerprints, source identifiers, repository identifiers, workflow paths, job and step names, and secret-variable names. These operations do not send application source code or the values of secrets as lockfile content. Review the generated file to understand the metadata associated with your project.
Request handling, account information, retention, hosting, and subprocessors are governed by the privacy policy and applicable service terms. This page does not claim that no operational logs or customer metadata are stored.
Understand what a signature establishes.
The platform signs scg.lock, and the CLI verifies it using the platform public key built into the binary. This protects the integrity of the signed record. It does not prove that your team approved a particular dependency or that the dependency is benign.
SCG relies on the CLI, its release channel, the platform, and the upstream observations available to the platform. Keep your own approval controls around changes to the baseline and the configuration that runs the check.
The installer compares the binary with a published checksum. Release signatures and available provenance are separate evidence that should be verified through the published release process. Do not assume that a checksum alone establishes who produced an artifact.
Need to evaluate a specific setup?
Tell us which operating systems, package managers, repositories, and build systems you use. We will discuss the supported boundary rather than assume your environment is covered.
Contact the SCG teamReport a vulnerabilityInspect the source