Agentic Actions Auditor

功能

• Detects prompt injection via direct expressions, env vars, and CLI fetches.
• Identifies dangerous sandbox configurations and wildcard user allowlists.
• Analyzes cross-file references for hidden AI agents.
• Provides detailed remediation guidance for each attack vector.
• Supports both local and remote repository analysis.

使用场景

• Auditing a repository's GitHub Actions workflows for AI agent security.
• Reviewing CI/CD configurations that invoke AI coding agents.
• Evaluating agentic action configurations for security risks.
• Assessing trigger events that expose workflows to external input.

非目标

• Performing runtime prompt injection testing.
• Auto-fixing or modifying workflow files.
• Auditing non-GitHub CI/CD systems.
• Reviewing standalone composite actions outside of a caller workflow context.

工作流

1. Determine analysis mode (local or remote).
2. Discover GitHub Actions workflow files.
3. Identify AI action steps using `uses:` references.
4. Capture security context (trigger events, env vars, permissions).
5. Analyze for specific attack vectors (env var intermediary, direct injection, etc.).
6. Report findings with detailed evidence, data flow, and remediation.

实践

• Security Auditing
• CI/CD Security
• Prompt Injection Detection
• Static Analysis

先决条件

• GitHub CLI (`gh`) installed and authenticated
• Access to the GitHub repository being audited