Don't feed the whole file to the AI: meet Sentinel-IR

Sentinel-IR is a free JavaScript package that condenses a source code file into a concise, structured note for language models. This note provides information about the file's purpose, potential risks, and specific details about its functionality without directly copying the source code.

The package is useful for both human users and developers working with AI coding tools. For humans, it offers a quick summary of the code without the need to sift through large amounts of text. For developers, it provides structured data that can be used to guide AI models in evaluating code changes, ensuring they only receive relevant information.

Sentinel-IR reads JavaScript files and generates notes containing essential facts such as routes, environment variables, operations, and potential risks. These notes are much smaller than the original files, often 40 times less, which reduces costs and speeds up processing. Importantly, the package does not rely on hidden instructions in code comments to influence AI behavior, making it a more reliable tool for AI-assisted code review.

Developed as part of an autonomous code management system called Sentinel, Sentinel-IR operates as an open-source tool under the Apache-2.0 license. While it cannot guarantee code safety or act as a security scanner, it significantly reduces the attack surface by limiting the information sent to AI models.