What problem does it solve?
This Skill helps you prevent prompt-injection and workflow exploitation risks in GitHub Actions that integrate agentic AI tooling, by statically auditing how untrusted inputs can reach AI prompts and execution paths.
Core Features & Use Cases
- Workflow discovery and AI step identification: Finds AI agent action steps across
.github/workflows and records where each AI integration is configured.
- Cross-file context tracing (depth-1): Resolves local composite actions and reusable workflows one level deep to uncover hidden AI agent usage.
- Attack vector detection for agent integrations: Detects high-risk patterns such as attacker-controlled event data injected into prompts (directly or via
env), runtime gh-based content fetch instructions, pull_request_target + PR-head checkout, dangerous sandbox/approval configurations, and wildcard allowlists.
- Actionable security reporting: Produces findings with severity, evidence snippets, data-flow traces, and remediation guidance aligned to supported action security profiles.
Quick Start
Use the agentic-actions-auditor skill to audit your repository’s .github/workflows for GitHub Actions that invoke AI coding agents and report prompt-injection and CI/CD security misconfigurations.
