Vorschlagen

Agentic Actions Auditor

Skill

Trail of Bits skill that audits GitHub Actions workflows wired to AI coding agents (Claude Code Action, Codex, Gemini CLI and others) for prompt injection, env-variable relays, dangerous sandbox settings, wildcard allowlists and other privilege risks.

trailofbits/skills · plugins/agentic-actions-auditor7.6k
Auf GitHub ansehen
README

agentic-actions-auditor

Audits GitHub Actions workflows for security vulnerabilities in AI agent integrations. Detects misconfigurations and attack vectors specific to Claude Code Action, Gemini CLI, OpenAI Codex, and GitHub AI Inference when used in CI/CD pipelines.

What It Does

This plugin provides a security audit skill that analyzes GitHub Actions workflow YAML files for vulnerabilities arising from AI agent integrations. It focuses on scenarios where attacker-controlled input (pull request titles, branch names, issue bodies, comments, commit messages, file contents, environment variables) can reach an AI agent running with elevated permissions in CI.

Attack Vectors Detected

The skill checks for nine categories of security issues:

  • A. Env Var Intermediary -- Attacker data flows through env: blocks to AI prompt fields with no visible ${{ }} expressions
  • B. Direct Expression Injection -- ${{ github.event.* }} expressions embedded directly in AI prompt fields
  • C. CLI Data Fetch -- gh CLI commands in prompts fetch attacker-controlled content at runtime
  • D. PR Target + Checkout -- pull_request_target trigger combined with checkout of PR head code
  • E. Error Log Injection -- CI error output or build logs fed to AI prompts carry attacker payloads
  • F. Subshell Expansion -- Restricted tools like echo allow subshell expansion (echo $(env)) bypass
  • G. Eval of AI Output -- AI response flows to eval, exec, or unquoted $() in subsequent steps
  • H. Dangerous Sandbox Configs -- danger-full-access, Bash(*), --yolo disable safety protections
  • I. Wildcard Allowlists -- allowed_non_write_users: "*" or allow-users: "*" permit any user to trigger

Supported AI Actions

ActionRepositoryNotes
Claude Code Actionanthropics/claude-code-action
Gemini CLIgoogle-github-actions/run-gemini-cliPrimary
Gemini CLI (legacy)google-gemini/gemini-cli-actionArchived
OpenAI Codexopenai/codex-action
GitHub AI Inferenceactions/ai-inference

Installation

From a project with the Trail of Bits internal marketplace configured:

/plugin menu

Select agentic-actions-auditor from the Security Tooling section.

Skills Included

SkillDescription
agentic-actions-auditorAudits GitHub Actions workflow files for AI agent security vulnerabilities

Target Audience

  • Security auditors reviewing repositories that use AI agents in CI/CD
  • Developers configuring Claude Code Action, Gemini CLI, OpenAI Codex, or GitHub AI Inference in their workflows
  • DevSecOps engineers establishing secure defaults for AI-assisted code review pipelines

Usage

Once installed, the skill activates automatically when Claude detects GitHub Actions workflow files (.github/workflows/*.yml) containing AI agent action references. You can also invoke it directly:

Audit the GitHub Actions workflows in this repository for AI agent security issues.

The skill produces a structured findings report covering each applicable attack vector with severity ratings and remediation guidance.

License

Creative Commons Attribution-ShareAlike 4.0 International

Ähnliche Ressourcen