What problem does it solve? AI coding agents running in CI/CD pipelines can receive attacker-controlled input through issue bodies, PR titles, and comments, enabling prompt injection, secret exfiltration, and remote code execution. This Skill statically audits GitHub Actions workflows to find these vulnerabilities before they are exploited. ## Core Features & Use Cases - AI Action Discovery: Identifies steps using Claude Code Action, Gemini CLI, OpenAI Codex, and GitHub AI Inference, including agents hidden inside composite actions and reusable workflows. - Nine Attack Vector Detections: Checks for env var intermediary injection, direct expression injection, CLI data fetches, pull_request_target checkout risks, error log injection, subshell expansion bypasses, eval of AI output, dangerous sandbox configs, and wildcard user allowlists. - Structured Findings Reports: Produces severity-rated findings with evidence snippets, data flow traces, and action-specific remediation guidance. - Use Case: Point the Skill at a repository like owner/repo and receive a report such as "Analyzed 5 workflows containing 3 AI action instances. Found 2 findings: 1 High, 1 Medium" with exact YAML evidence and fixes. ## Quick Start Audit the GitHub repository owner/repo for prompt injection vulnerabilities in any workflows that invoke AI coding agents.
