Statically scan agent repos for prompt injection and unsafe MCP configs with Agent Audit

Audit agent code, prompts, and MCP configuration for prompt-injection surfaces, taint issues, and unsafe tool exposure before shipping.

Prerequisites

agent-audit, local agent repository or config tree

Installation

Use the upstream install or setup path that matches your environment:

• pip install agent-audit
• git clone https://github.com/HeadyZhang/agent-audit

Requirements and caveats from upstream:

• 
• | T6 | openai-agents-python | 25 | ASI-01, ASI-02 |
• | T7 | adk-python | 40 | ASI-02, ASI-04, ASI-10 |

Basic usage or getting-started notes:

• Install

• bash

• Scan your project

• Source: https://github.com/HeadyZhang/agent-audit

• Extracted from upstream docs: https://raw.githubusercontent.com/HeadyZhang/agent-audit/HEAD/README.md

Documentation

• https://headyzhang.github.io/agent-audit/

Source

• Agent Skill Exchange