Kereva: Static analysis for LLM-powered Python applications

by ddos · June 18, 2025

Kereva LLM Code Scanner is a static analysis tool designed to identify potential security risks, performance issues, and vulnerabilities in Python codebases that use Large Language Models (LLMs). It analyzes your code without execution to detect problems like hallucination triggers, bias potential, prompt injection vulnerabilities, and inefficient LLM usage patterns.

LLM Security, Code Scanner

Features

Static Code Analysis: Find issues without executing your code
Specialized LLM Scanners: Detect security, quality, and efficiency problems specific to LLM applications
Multi-format Support: Analyze Python files and Jupyter notebooks (.ipynb)
Flexible Reporting: Get results in human-readable console output or structured JSON

Scanner Categories

Prompt Scanners (`prompt`)

Scanner	Description
`xml_tags`	Detects improper usage of XML tags in prompts
`subjective_terms`	Identifies undefined subjective assessments
`long_list`	Flags attention issues with large data lists
`inefficient_caching`	Locates inefficient prompt caching patterns
`system_prompt`	Checks for missing or misplaced system prompts in LLM API calls

Chain Scanners (`chain`)

Scanner	Description
`unsafe_input`	Identifies unsanitized user input flowing to LLMs
`langchain`	Detects LangChain-specific vulnerabilities
`unsafe_output`	Detects vulnerabilities where LLM output is used without proper sanitization in security-sensitive operations

Output Scanners (`output`)

Scanner	Description
`unsafe_execution`	Finds potential code execution risks from LLM outputs
`structured`	Validates output model definitions and constraints
`unsafe_rendering`	Detects when LLM output is used in rendering functions without proper sanitization
`safe_shell_commands`	Enforces safe shell command execution with LLM outputs
`huggingface_security`	Identifies security vulnerabilities in HuggingFace model usage