🛠 Instructions in AGENTS.md can lie to your AI agents
The industry is moving toward managing AI agents through files in repositories (AGENTS.md, CLAUDE.md), but text instructions quickly become outdated. Project Hunch has released version v1.1, which links Markdown to an architectural decision graph. Using special HTML comments, text can be tied to specific decisions, allowing CI systems (via the hunch drift command) to block changes if instructions contradict the current state of the architecture.
🌍 This solves the problem of "documentation drift" in the era of AI agents. Instead of passive reading, documentation becomes an active participant in the quality control process, where discrepancies between code and rules are prevented at the CI/CD stage.
👤 This is a way to ensure that your AI assistant won't suggest outdated or incorrect architectural patterns simply because they are still described in the README or AGENTS.md.
Source 1: https://hunch-pi.vercel.app/blog/post?slug=agents-md-is-lying-to-your-agent