agentleFS
nitinjain999nitinjain999/platform-skills/skills/supply-chain/SKILL.md

supply-chain

Skill71 starsChanged 8 days agoApache-2.0View on GitHub

Secure the software supply chain from source to running container. Covers Cosign keyless image signing (Sigstore/Rekor), SBOM generation and attestation (Syft), vulnerability scanning with severity gates (Trivy/Grype), SLSA Level 2 provenance, and Kyverno/OPA admission enforcement. All open-source, no license cost. Use when asked to "sign my image", "generate an SBOM", "scan for CVEs", "attest build provenance", "enforce image signatures in Kubernetes", or "implement SLSA".

  • Reads credentials

We split this skill into the rules it gives an agent

9 rules. 0 are common, 8 are distinctive, and 1 is very specific to this project.

  • 0

    Common

    Most of the 7 edited versions we read keep these.

  • 8

    Distinctive

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    When invoked with no arguments, ask before proceeding:

    line 16
  2. Distinctive:

    Ask for or read: the GitHub Actions workflow file(s), any existing image scanning steps, and any Kyverno/OPA admission policies

    line 48
  3. Distinctive:

    Never store signing keys in CI secrets — use keyless only

    line 91
  4. Distinctive:

    Always sign the digest, not the tag

    line 92
  5. Distinctive:

    Add the SBOM generation step after build and push (digest is only available after registry push):

    line 102
  6. Distinctive:

    Add a CVE vulnerability scan with a configurable severity gate that fails the build.

    line 130
  7. Distinctive:

    Add Trivy scan step:

    line 133
  8. Distinctive:

    Ask: keyless (Sigstore) or key-based signing?

    line 166
  9. Very specific:

    After completing this task, log errors and learnings via /platform-skills:self-improve log.

    line 257

“Common” is measured: kept by at least 6 in 10 of the 7 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Secure the software supply chain — from the build pipeline to running containers.


Interactive Wizard (fires when no arguments are provided)

When invoked with no arguments, ask before proceeding:

Q1 — Mode?

What do you need?
  1. audit    — review an existing CI/CD pipeline for supply chain security gaps
  2. sign     — set up keyless image signing with Cosign (Sigstore/Rekor, no key management)
  3. sbom     — generate and attest an SBOM with Syft
  4. scan     — add a CVE vulnerability gate with Trivy or Grype
  5. enforce  — write a Kyverno policy to block unsigned images at admission
  6. slsa     — generate a SLSA Level 2 provenance workflow

Enter 1–6 or mode name:

Q2 — Context (after mode selected, one at a time):

  • audit: Paste your GitHub Actions workflow file(s) or describe your current CI/CD pipeline:
  • sign: Which container registry? (ECR / GHCR / Docker Hub / other):
  • sbom: Which image and registry? (e.g. ghcr.io/org/image):
  • scan: Which scanner preference? Trivy (recommended) or Grype — or no preference:
  • enforce: Which registry or image prefix should the policy cover? (e.g. ghcr.io/myorg/*):
  • slsa: Which registry and repo? (e.g. ghcr.io/org/image from github.com/org/repo):

Then proceed into the relevant mode below.


Mode: audit

Review an existing CI/CD pipeline and cluster admission configuration for supply chain security gaps.

Steps:

  1. Ask for or read: the GitHub Actions workflow file(s), any existing image scanning steps, and any Kyverno/OPA admission policies
  2. Classify gaps by severity:
    • Critical: images pushed without signing, no CVE gate, unsigned images admitted to cluster
    • High: SBOM not generated or not attested, severity gate only on CRITICAL (misses HIGH)
    • Medium: action versions pinned to tag not SHA, no SLSA provenance
  3. Output a prioritised gap list:
    [CRITICAL] No image signing — any image can be admitted to the cluster
    [CRITICAL] No CVE severity gate — vulnerable images pass CI
    [HIGH]     No SBOM attestation — cannot audit what is running
    [MEDIUM]   Action versions pinned to tag, not SHA
  4. Recommend the fix order: sign → scan-gate → SBOM → enforce → SLSA

Reference: references/supply-chain.md → Gap classification, Fix order

Mode: sign

Set up keyless image signing with Cosign using Sigstore/Rekor (no key management required).

Steps:

  1. Confirm the image registry (ECR, GHCR, Docker Hub) and CI platform (GitHub Actions assumed)
  2. Generate the signing workflow step:
    - name: Install Cosign
      uses: sigstore/cosign-installer@11086d9f32b178aa24e93c2b86eba3ef4b16b68a  # v3.8.1
    
    - name: Sign image
      run: |
        cosign sign --yes \
          ${{ env.REGISTRY }}/${{ env.IMAGE_NAME }}@${{ steps.build.outputs.digest }}
  3. Explain the keyless flow: GitHub Actions OIDC token → Fulcio CA → Rekor transparency log
  4. Show verification command:
    cosign verify \
      --certificate-identity-regexp="https://github.com/<org>/<repo>/.github/workflows/.*" \
      --certificate-oidc-issuer="https://token.actions.githubusercontent.com" \
      ghcr.io/<org>/<image>@<digest>
  5. Flag: always sign the digest (@sha256:…), never the tag — tags are mutable

Key rules:

  • Never store signing keys in CI secrets — use keyless only
  • Always sign the digest, not the tag
  • Pin cosign-installer to a full SHA, e.g.: sigstore/cosign-installer@11086d9f32b178aa24e93c2b86eba3ef4b16b68a

Reference: references/supply-chain.md → Keyless signing, Rekor transparency log

Mode: sbom

Generate a Software Bill of Materials with Syft and attest it as an OCI artifact alongside the image.

Steps:

  1. Add the SBOM generation step after build and push (digest is only available after registry push):
    - name: Generate SBOM
      uses: anchore/sbom-action@61119d458adab75f756bc0b9e4bde25725f86a7a  # v0.20.0
      with:
        image: ${{ env.REGISTRY }}/${{ env.IMAGE_NAME }}@${{ steps.build.outputs.digest }}
        format: spdx-json
        output-file: sbom.spdx.json
    
    - name: Attest SBOM
      run: |
        cosign attest --yes \
          --predicate sbom.spdx.json \
          --type spdxjson \
          ${{ env.REGISTRY }}/${{ env.IMAGE_NAME }}@${{ steps.build.outputs.digest }}
  2. Show how to retrieve the SBOM attestation:
    cosign download attestation \
      --predicate-type https://spdx.dev/Document \
      ghcr.io/<org>/<image>@<digest> | jq '.payload | @base64d | fromjson'
  3. Note: use spdx-json format for broadest tooling compatibility; cyclonedx-json is an alternative

Reference: references/supply-chain.md → SBOM formats, Syft, Attestation

Mode: scan

Add a CVE vulnerability scan with a configurable severity gate that fails the build.

Steps:

  1. Add Trivy scan step:
    - name: Scan image for vulnerabilities
      uses: aquasecurity/trivy-action@18f2135c0b15d26b3a4c2efded75e06b6f0e4884  # v0.30.0
      with:
        image-ref: ${{ env.REGISTRY }}/${{ env.IMAGE_NAME }}@${{ steps.build.outputs.digest }}
        format: table
        exit-code: '1'
        ignore-unfixed: true
        vuln-type: os,library
        severity: CRITICAL,HIGH
  2. Explain ignore-unfixed: true — skip CVEs with no fix available (reduces noise without reducing security)
  3. For Grype (alternative), show:
    - name: Scan with Grype
      uses: anchore/scan-action@e1165082ffb1fe366ebaf02d8526e7c4989ea9d2  # v7.4.0
      with:
        image: ${{ env.REGISTRY }}/${{ env.IMAGE_NAME }}@${{ steps.build.outputs.digest }}
        fail-build: true
        severity-cutoff: high
  4. Recommend Trivy for new setups (CNCF, single binary, SBOM support); Grype for teams already using Syft

Reference: references/supply-chain.md → Trivy vs Grype, Severity gates

Mode: enforce

Generate a Kyverno ImageValidatingPolicy that blocks unsigned or unverified images at admission time.

Finding where unsigned images already run: Kubescape reports them as control C-0237 "Check if signature exists" (/platform-skills:kubescape-operator, or kubescape scan control C-0237 on a cluster). Use that list to size the rollout before switching the policy to Deny.

Steps:

  1. Ask: keyless (Sigstore) or key-based signing?
  2. For keyless, generate:
    apiVersion: policies.kyverno.io/v1
    kind: ImageValidatingPolicy
    metadata:
      name: require-signed-images
    spec:
      validationActions: [Audit]   # switch to [Deny] after all images are signing
      matchConstraints:
        resourceRules:
        - apiGroups: [""]
          apiVersions: ["v1"]
          operations: ["CREATE", "UPDATE"]
          resources: ["pods"]
      matchImageReferences:
      - glob: "ghcr.io/<org>/*"
      validations:
      - expression: >
          images.containers.map(image, verifyImageSignatures(image, [{"keyless": {
            "url": "https://fulcio.sigstore.dev",
            "rekor": {"url": "https://rekor.sigstore.dev"},
            "identities": [{"issuer": "https://token.actions.githubusercontent.com",
              "subjectRegExp": "https://github.com/<org>/.*"}]
          }}])).all(e, e > 0)
        message: "Image must be signed via Sigstore keyless signing from GitHub Actions"
  3. Show audit-first deployment:
    spec:
      validationActions: [Audit]   # start here, switch to [Deny] after validation
  4. Cross-reference: /platform-skills:kyverno for full ImageValidatingPolicy guidance

Reference: references/supply-chain.md → Kyverno enforcement, references/kyverno.md → ImageValidatingPolicy

Mode: slsa

Generate a GitHub Actions workflow for SLSA Level 2 provenance using slsa-github-generator.

Steps:

  1. Explain what SLSA L2 gives: signed provenance linking the artifact to the specific build inputs (source commit, workflow, runner)
  2. Generate the workflow using the official reusable workflow:
    jobs:
      build:
        runs-on: ubuntu-latest
        outputs:
          digest: ${{ steps.build.outputs.digest }}
        permissions:
          contents: read
          packages: write
        steps:
          - name: Build and push image
            id: build
            uses: docker/build-push-action@48aba3b46d1b1fec4febb7c5d0c644b249a11355  # v6.10.0
            with:
              push: true
              tags: ghcr.io/<org>/<image>:${{ github.sha }}
    
      provenance:
        needs: build
        permissions:
          actions: read
          id-token: write
          packages: write
        uses: slsa-framework/slsa-github-generator/.github/workflows/generator_container_slsa3.yml@5a775b367a56d5bd118a224a811bba288150a563  # v2.0.0
        with:
          image: ghcr.io/<org>/<image>
          digest: ${{ needs.build.outputs.digest }}
          registry-username: ${{ github.actor }}
        secrets:
          registry-password: ${{ secrets.GITHUB_TOKEN }}
  3. Show provenance verification:
    slsa-verifier verify-image \
      ghcr.io/<org>/<image>@<digest> \
      --source-uri github.com/<org>/<repo> \
      --source-branch main
  4. Note: generator_container_slsa3.yml produces L2 provenance by default; L3 requires hermetic builds not available on standard GitHub-hosted runners

Reference: references/supply-chain.md → SLSA levels, slsa-github-generator


Signing and SBOMs cover what the pipeline produces. They say nothing about the pipeline itself — an unpinned uses: or a pull_request_target checkout of untrusted code compromises the build before any artifact exists. To audit the workflows that run this supply chain, including unpinned-uses hash-pin policy enforcement, use /platform-skills:zizmor. They also say nothing about whether a credential leaked into the repo that produces this artifact in the first place — for that, live-validate and revoke with /platform-skills:kingfisher.


After completing this task, log errors and learnings via /platform-skills:self-improve log.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.