Imported from personamanagmentlayer/pcl (stdlib/security/supply-chain-security-expert/SKILL.md). Install upstream with npx skills add personamanagmentlayer/pcl --skill supply-chain-security-expert. Copyright stays with the author.

Supply Chain Security Expert

Most of what ships is code nobody on the team wrote. OWASP ranks software supply chain failures among the highest-impact categories in the 2025 Top 10, and EU regulation (NIS2, and the Cyber Resilience Act) now makes dependency management a compliance obligation rather than an engineering preference.

Core Concepts

The Attack Surface

Know What You Ship

You cannot patch what you cannot enumerate. An SBOM is the inventory — generated at build time from what actually went in, not written by hand.

Provenance Is Stronger Than Scanning

Scanning tells you about known vulnerabilities. Provenance tells you that this artefact was built from this commit by that pipeline. When a new advisory lands, scanning finds it; when a build system is compromised, only provenance does.

SBOM

Generate at build time, store as a build artefact, attach to the release.

# From a container image, a filesystem, or a lockfile
syft packages ghcr.io/acme/api:1.4.2 -o cyclonedx-json > sbom.cdx.json
syft packages dir:. -o spdx-json > sbom.spdx.json

# Scan the SBOM rather than re-resolving dependencies
grype sbom:sbom.cdx.json --fail-on high
osv-scanner --sbom sbom.cdx.json

CycloneDX suits security use; SPDX suits licence compliance. Producing both costs one extra command.

An SBOM is only useful if it is queryable when an advisory lands:

# "Are we affected by the log4j-style advisory in package X?"
jq -r '.components[] | select(.name == "log4j-core") | "\(.name)@\(.version)"' sbom.cdx.json

Store SBOMs per release, indexed by artefact digest, retained as long as the artefact runs anywhere. The question always arrives after the release.

Dependency Hygiene

Pin, and pin transitively

npm ci                              # lockfile only; fails if package.json disagrees
pip install --require-hashes -r requirements.txt
cargo build --locked
go mod verify

npm install in CI can resolve differently from local. npm ci cannot — it is the difference between a reproducible build and a hope.

# requirements.txt with hashes: the registry cannot substitute content
cryptography==43.0.1 \
    --hash=sha256:8d09d05439ce7baa8e9e95b07ec5b6c886f548deb7e0f69ef25f64b3bce842f2

Review what you add

Before a new direct dependency: is it maintained, does it have more than one maintainer, what does it pull in transitively, what licence, and does it run install scripts?

npm view <pkg> maintainers time.modified
npm ls <pkg> --all                          # who actually depends on it
npx license-checker --summary
docker run --rm -v "$PWD:/src" aquasec/trivy fs --scanners vuln,license /src

Disable install scripts by default and allow them explicitly:

# .npmrc
ignore-scripts=true

Most package-manager compromises execute at install time. This one line removes that class.

Prevent dependency confusion

# .npmrc — internal scope resolves only from the internal registry
@acme:registry=https://npm.internal.acme.com/
//npm.internal.acme.com/:_authToken=${NPM_TOKEN}

# pip.conf — never mix; --extra-index-url silently prefers the highest version
[global]
index-url = https://pypi.internal.acme.com/simple

--extra-index-url pointing at both an internal and the public index is the dependency confusion vulnerability, not a mitigation: pip picks the highest version wherever it lives.

Vulnerability Management

Scanning produces more findings than anyone can fix. Triage by exploitability, not by CVSS alone.

osv-scanner --lockfile package-lock.json --lockfile requirements.txt
grype ghcr.io/acme/api:1.4.2 --fail-on critical --only-fixed
trivy image --severity HIGH,CRITICAL --ignore-unfixed ghcr.io/acme/api:1.4.2

Triage questions, in order:

1. Is the vulnerable code path reachable from our application? Reachability analysis eliminates the majority of findings.
2. Is it exploitable in our deployment — is that parser fed untrusted input?
3. Is a fix available? --only-fixed separates actionable from noise.
4. What is the exposure — internet-facing, or an internal batch job?

Record the decision. A VEX document states, machine-readably, that a component is present but not affected:

{
  "bomFormat": "CycloneDX",
  "specVersion": "1.6",
  "vulnerabilities": [
    {
      "id": "CVE-2026-12345",
      "analysis": {
        "state": "not_affected",
        "justification": "vulnerable_code_not_in_execute_path",
        "detail": "XML parsing is not reached; we use the JSON codec exclusively."
      }
    }
  ]
}

Without VEX, the same finding is re-triaged by every scanner run and every customer questionnaire.

Build Integrity

SLSA describes build integrity in levels. The useful progression:

name: release
on:
  push:
    tags: ['v*']

permissions:
  contents: read
  id-token: write # OIDC: no long-lived registry credentials
  attestations: write
  packages: write

jobs:
  build:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
      - name: Build and push
        id: push
        run: |
          docker build -t "$IMAGE:$GITHUB_REF_NAME" .
          docker push "$IMAGE:$GITHUB_REF_NAME"
          echo "digest=$(docker inspect --format='{{index .RepoDigests 0}}' "$IMAGE:$GITHUB_REF_NAME")" >> "$GITHUB_OUTPUT"

      - uses: actions/attest-build-provenance@v2
        with:
          subject-name: ${{ env.IMAGE }}
          subject-digest: ${{ steps.push.outputs.digest }}
          push-to-registry: true

      - name: Generate and attach SBOM
        run: |
          syft "$IMAGE@${{ steps.push.outputs.digest }}" -o cyclonedx-json > sbom.cdx.json
          cosign attest --predicate sbom.cdx.json --type cyclonedx \
                        "$IMAGE@${{ steps.push.outputs.digest }}"

Pin actions to a commit SHA, not a tag. A tag is mutable: @v4 can be moved to point at anything, and that has been used in real attacks.

Signing and Verification

Signing that is never verified provides nothing. The verification step is the control.

# Keyless signing: identity from the OIDC token, recorded in a public transparency log
cosign sign --yes ghcr.io/acme/api@sha256:abc123...

# Verification, bound to the workflow that is allowed to build this
cosign verify ghcr.io/acme/api@sha256:abc123... \
  --certificate-identity-regexp="^https://github.com/acme/api/.github/workflows/release.yml@refs/tags/v" \
  --certificate-oidc-issuer=https://token.actions.githubusercontent.com

The identity constraints are the security. cosign verify without them confirms only that somebody signed it.

Enforce at admission, so an unsigned image cannot run:

apiVersion: kyverno.io/v1
kind: ClusterPolicy
metadata:
  name: require-signed-images
spec:
  validationFailureAction: Enforce
  rules:
    - name: verify-signature
      match:
        any: [{ resources: { kinds: [Pod] } }]
      verifyImages:
        - imageReferences: ['ghcr.io/acme/*']
          attestors:
            - entries:
                - keyless:
                    subject: 'https://github.com/acme/*/.github/workflows/release.yml@refs/tags/*'
                    issuer: 'https://token.actions.githubusercontent.com'

CI/CD Hardening

• Least-privilege tokens. Default permissions: {} at the workflow level and grant per job.
• No long-lived cloud keys. OIDC federation instead.
• Never run untrusted code with secrets in scope. pull_request_target with a checkout of the fork's head is a well-known privilege escalation.
• Pin actions and base images by digest.
• Separate build from deploy credentials. A compromised build job should not be able to deploy.
• Protect the release branch: required reviews, no force push, signed commits.
• Log and retain build metadata — who triggered it, from which commit.

Best Practices

• Generate an SBOM for every build, retained with the artefact.
• Scan continuously, not only at build. New advisories apply to artefacts already released.
• Fail the build on new critical findings, and allow documented exceptions with an expiry date.
• Automate dependency updates (Renovate, Dependabot) with grouped PRs and a cooldown before adopting a brand-new release.
• Keep an inventory of what runs where, so an advisory maps to deployments in minutes rather than days.
• Prefer fewer dependencies. Every one is a trust decision and a maintenance obligation.

Anti-Patterns

• Floating versions (^1.2.3, latest) in production builds.
• npm install in CI instead of npm ci.
• Tag-pinned actions and base images — mutable references.
• Signing without verifying.
• Scanning only at build time.
• Ignoring transitive dependencies — most vulnerable code is transitive.
• --extra-index-url mixing internal and public registries.
• An SBOM generated once for an audit and never again.
• Blanket suppression of findings with no justification or expiry.

Reference Documentation

• Response Playbook — responding to a compromised dependency, incident triage, VEX authoring, and regulatory obligations

Resources

• SLSA framework
• Sigstore and cosign
• CycloneDX and SPDX
• OSV database and scanner
• OpenSSF Scorecard
• CISA SBOM guidance