Open-Source Proxy Tool Blocks AI Coding Agents from Leaking Secrets to the Cloud

AI coding agents like Claude Code, Cursor, and Codex can read local files and terminal output, meaning sensitive credentials in files like .env may be inadvertently sent to cloud model providers as part of prompts. A developer has built Anonmyz, an open-source local proxy tool designed to intercept outbound requests from AI coding agents before they leave the machine. The tool scans request payloads for secrets, replaces them with unique cryptographic placeholders, and restores original values locally once the model response returns. Unlike redaction, the placeholder approach lets the model reason about data structure without ever seeing real credentials. Anonmyz also handles streaming responses, where secrets or placeholders can be split across network chunks, using a look-behind window to ensure nothing slips through.