Source-control trust
Branch protection, required reviews, signed commits, GitHub Apps and PATs scope, dependency-bot trust, third-party Action trust.
build-system audit · 10
Audit the path from commit to production.
Dependency trust, build-system integrity, secret handling in CI, artifact signing, deployment-credential blast radius, GitOps trust chains. Every place an attacker could ship code into your production environment without writing a line of yours.
01. in scope
What's in scope.
What we audit in a supply-chain engagement.
Build-system integrity
CI runner isolation, secrets injection, cache poisoning risk, build provenance, reproducibility of artifacts.
Dependency supply chain
Direct and transitive dependencies, typosquatting risk, package signing, lockfile integrity. Coverage of npm, PyPI, Maven, Go modules, Cargo, NuGet, RubyGems.
Container and image supply
Base-image trust, layer provenance, vuln scanning at build vs run, registry signing, deploy-key blast radius.
Artifact signing and verification
Sigstore / cosign / SLSA adoption, verification at admission (Kyverno / OPA / native), key rotation policy.
GitOps and CD trust
ArgoCD / Flux trust model, deployment-credential scope, rollback authorization, environment promotion gates.
02. how we work
How we work on it.
How a supply-chain audit runs.
- Scoping call60 minutes. We learn the platform: source-control vendor, CI vendor, CD model, runtime, registry.
- Configuration reviewWe read the YAML, the policies, the access lists, the audit logs. Discover what is actually configured, not what the doc says.
- Active validationWe try the realistic attack paths. PR poisoning, dependency-confusion, secret exfiltration via CI, malicious Action via fork.
- Report + readoutFindings with attack chains, fix steps, suggested CI rules and admission policies.
- Retest30 days, one round, included. Confirms the policies stick after deploy.
03. deliverables
What you walk away with.
Supply-chain audit deliverables.
Trust-chain map
Diagram from developer commit to running container. Every trust boundary marked. Useful for next year's audit too.
Findings with attack chains
Each finding shows the exploit path, not just the misconfiguration. Severity reflects blast radius.
CI policy recommendations
Concrete actions for branch protection, required reviews, secret-scope reduction, build-runner isolation.
SLSA / Sigstore adoption plan
If you want to harden further: realistic adoption plan with effort estimate and what each level buys.
Admission-policy starter pack
OPA / Kyverno / native policies for image signing, registry trust, and deploy gating.
Developer-facing security guide
A short doc your engineers actually read. What to do, what to avoid, who to ask when the workflow blocks.
04. when
When teams hire us for this.
When this audit matters.
You use a lot of GitHub Actions or GitLab CI
Action ecosystems are increasingly the attack surface. Third-party Actions run with your secrets. We map the trust and tighten it.
You ship containers
Container supply chain has more moving parts than most teams realize. Base images, layer caches, registries, admission policy.
You adopted GitOps recently
GitOps shifts the trust boundary from CD pipeline to repo. New attack surface; audit it.
You need SLSA evidence
Customer or government ask. SLSA Level 2 or 3 is the new baseline expectation for enterprise sales.
05. faq
Questions before the call.
Supply-chain FAQ.
Is this a SBOM audit?
SBOM is part of it, not the whole. We care about the trust chain, not just the package list. SBOM accuracy and currency are checked.
Do you require admin access to GitHub / GitLab?
No. Org-reader and CI-reader is enough for most findings. Admin needed only if you want us to implement the recommendations.
What if we use a self-hosted runner?
We audit it. Self-hosted runners are a common compromise path, especially when shared across repos.
What about secrets in code?
We scan and triage. Most teams have at least one. We document, recommend rotation, and write a detection rule for the next time.
Do you cover the model artifact pipeline (AI workflows)?
Yes. Model weights, training data, evaluation datasets are increasingly part of the supply chain. Ask if your AI features should be in scope.
Want to know if your build pipeline would survive a real attack?
60-minute call covers your stack and the most likely top-three findings before we quote.