Agent Skills: Third-Party Action Risk Assessment

>-

UncategorizedID: adaptive-enforcement-lab/claude-skills/third-party-action-risk-assessment

Install this agent skill to your local

pnpm dlx add-skill https://github.com/adaptive-enforcement-lab/claude-skills/tree/HEAD/plugins/secure/skills/third-party-action-risk-assessment

Skill Files

Browse the full folder contents for third-party-action-risk-assessment.

Download Skill

Loading file tree…

plugins/secure/skills/third-party-action-risk-assessment/SKILL.md

Skill Metadata

Name
third-party-action-risk-assessment
Description
>-

Third-Party Action Risk Assessment

When to Use This Skill

Trust but verify. Every third-party action you adopt into your workflows executes with access to your secrets, code, and deployment infrastructure. Know what you're trusting.

The Risk

Third-party actions run arbitrary code inside your CI/CD pipeline with full access to repository secrets, cloud credentials, and source code. A malicious or compromised action can exfiltrate everything, deploy backdoors, or modify your codebase.

Implementation

See the full implementation guide in the source documentation.

Key Principles

Always SHA pin third-party actions: Tag references can be mutated. SHA pins are immutable.

# Bad - tag reference
- uses: community/action@v2

# Good - SHA pinned with version comment
- uses: community/action@a1b2c3d4e5f6...  # v2.1.0

Review action source code before first use: Never trust based on stars or README alone. Read the actual implementation.

Fork critical actions to organization control: Removes dependency on external maintainer. Gives you control over updates.

Monitor for action updates: Use Dependabot to track new versions. Review changelogs before updating.

Minimize permissions: Grant actions only what they need. Use job-level scoping to limit scope.

Isolate high-risk workflows: Run untrusted actions in separate jobs with minimal permissions and no secret access.

Audit action usage quarterly: Review which actions are in use. Re-assess risk as threat landscape evolves.

Have an exit strategy: Know how to replace or remove every action if it becomes compromised or unmaintained.

Examples

See examples.md for code examples.

Full Reference

See reference.md for complete documentation.

References