agentic-actions-auditor

To audit GitHub Actions workflows for AI agent prompt injection risks, use static analysis to trace attacker-controlled input from PRs or issues to AI prompts and detect dangerous env var intermediary patterns and subshell expansion bypasses in your CI/CD pipeline configuration.

An env var intermediary pattern in CI/CD security is an attack vector where attacker-controlled input is passed through environment variables to AI agents, creating prompt injection risks that allow unauthorized shell command execution or malicious prompt manipulation in GitHub Actions workflows.

Detect dangerous sandbox configurations in GitHub Actions by performing static security analysis on workflow files to identify overly permissive tool allowlists and trace AI agent usage through composite actions and reusable workflows for hidden security vulnerabilities.

Yes, static analysis can automatically trace AI agent usage through composite actions and reusable workflows via cross-file resolution to uncover hidden security vulnerabilities and direct expression injection risks in your CI/CD pipeline configurations.

Run a security audit on CI/CD pipelines using AI coding agents before merging new workflows to ensure attacker-controlled input from PRs or issues cannot reach the AI prompt or trigger unauthorized shell commands through dangerous sandbox configurations.

Static analysis for detecting prompt injection in GitHub Actions is limited to identifying known dangerous patterns in workflow files, env var intermediaries, and subshell expansion bypasses, and may not catch dynamically generated code or runtime-only prompt injection vectors.