| name | supply-chain-hardening |
| description | Use this skill when the user asks to "harden supply chain", "secure dependencies", "pin versions", "audit packages", "secure GitHub Actions", "harden containers", "secure IaC", "audit extensions", "secure AI models", "audit pickle files", "scan for secrets", "harden credentials", "rotate tokens", "audit credentials", "sign commits", "commit signing", "signed commits", "gitsign", or needs guidance on software supply chain security, dependency management, credential hygiene, secret scanning, or preventing supply chain attacks. Also trigger when the user mentions tools like Grype, Cosign, Sigstore, Checkov, Hadolint, Zizmor, ModelScan, SafeTensors, Betterleaks, gitsign, or references SLSA.
|
| version | 0.2.0 |
Supply Chain Security Hardening
Secure the software supply chain across seven domains: third-party packages, container images, GitHub Actions, infrastructure-as-code, AI/ML models, IDE extensions, and credentials/secrets.
How to Use This Skill
This skill provides two modes of operation:
Action Commands - Take Realistic Action by Default
These commands scan the codebase, make changes directly (pin hashes, fix configs, resolve SHAs), and explain each change. They are the default way to harden a project.
/audit-supply-chain - Full audit with auto-fixes for critical items/harden-packages - Pin versions, disable scripts, secure registries/harden-containers - Pin digests, enforce non-root, add .dockerignore/harden-actions - Pin SHAs, set permissions, fix injection, add Dependabot/harden-iac - Pin modules/providers, generate lockfiles, flag provisioners/harden-ai-ml - Fix unsafe deserialization, pin model sources/harden-ide-extensions - Audit extensions, remove secrets, add devcontainer/harden-credentials - Scan for leaked secrets, set up pre-commit hooks, harden credential hygiene/audit-credentials - Find long-lived tokens, hardcoded secrets, credentials to rotate or replace/update-pins - Check all pinned deps/actions/images for newer versions and update them/minimize - Remove unused dependencies and convert Dockerfiles to multi-stage builds
Walkthrough Commands - Guided Setup for Advanced Items
These commands are interactive walkthroughs for configurations that require steps outside the codebase (cloud provider setup, Kubernetes config, GitHub settings). They should be run separately.
/setup-oidc - Replace cloud credentials with OIDC tokens/setup-image-signing - Sign images with Cosign/Sigstore/setup-tag-rulesets - Protect tags from force-push attacks/setup-admission-control - Enforce image policies in Kubernetes/setup-sbom - Generate SBOMs and provenance attestations/setup-commit-signing - Set up commit and tag signing (SSH, GPG, or gitsign)/setup-runner-monitoring - Add runtime detection to CI runners
Reference Material
- Read the checklist at
references/checklist.md to understand what needs to be done
- Read the tool reference at
references/tools.md for open source and paid options
- Read the implementation configs at
references/package-configs.md for copy-paste configurations
- Read the container guidance at
references/container-configs.md for Dockerfile and image hardening
- Read the actions guidance at
references/actions-configs.md for GitHub Actions hardening
- Read the iac guidance at
references/iac-configs.md for Terraform/IaC hardening
- Read the credential guidance at
references/credentials-configs.md for secret scanning and credential hygiene
- Read the AI/ML guidance at
references/ai-ml-configs.md for model security configurations
Core Principles
- Pin everything. Versions, digests, commit SHAs. Floating references are attack surface.
- Never fabricate pins. Resolve deterministically. Always resolve versions, SHAs, and digests from live sources. Use methods that produce the same result regardless of local platform or environment:
docker buildx imagetools inspect for multi-arch manifest digests (not docker pull/docker inspect which are platform-specific), git ls-remote refs/tags/{tag}^{} to dereference annotated tags to commit SHAs (not the tag object SHA). If resolution fails, add a # TODO: pin comment with the exact command the user should run — never guess a value.
- Least privilege everywhere. Tokens, permissions, workflow scopes, container users.
- Verify provenance. Sign artifacts, check signatures, require attestations.
- Detect at runtime. Scanning at build time isn't enough - monitor what actually runs.
- Plan for cascading compromise. One stolen token can hit npm, PyPI, Docker Hub, and extension marketplaces simultaneously.
Domain Quick Reference
Third-Party Packages
Pin versions. Require cooldown periods. Disable install scripts. Use a registry proxy/firewall. Scan for malware. Generate SBOMs. See references/package-configs.md for language-specific configurations.
Container Images
Pin by digest. Run as non-root. Use minimal base images. Sign with Cosign. Scan continuously. Use admission controllers. See references/container-configs.md.
GitHub Actions
Pin to commit SHAs. Set explicit permissions. Audit for pull_request_target. Enable tag protection rules. Monitor runner activity. See references/actions-configs.md.
Infrastructure-as-Code
Pin module versions. Scan with Checkov/tflint. Enforce policy with OPA/Sentinel. Lock down provisioners. Require signed commits. See references/iac-configs.md.
AI/ML Models
Never load untrusted pickles. Use SafeTensors/ONNX. Verify model hashes. Scan with Picklescan/ModelScan. Sandbox model loading.
IDE Extensions
Audit installed extensions. Use osquery for fleet visibility. Enforce allowlists. Use VS Code Profiles. Run dev environments in containers.
Credentials & Secrets
Scan for leaked secrets with Betterleaks. Set up pre-commit hooks to prevent future leaks. Harden .gitignore. Rotate compromised credentials. Replace long-lived tokens with OIDC where possible. Sign commits to prevent impersonation. See references/credentials-configs.md.
When Making Changes
Action commands follow these principles:
- Make changes by default - pin versions, resolve SHAs, fix configs directly
- Explain each change as it's made - what was changed and why
- Don't break existing functionality - test after changes
- Prioritize "Do Right Now" items from the checklist before "Do Eventually" items
- Flag items that need manual attention separately from automated fixes
- Point users to walkthrough commands for advanced items that require external configuration
