AI-agent instructions become stale as the repo changes?

[raushankcode]

Hey, I’m asking a product-research question, not trying to promote blindly.

I noticed many TypeScript/AI/dev-tool repos now use files like AGENTS.md, CLAUDE.md, or Cursor rules to guide coding agents.

I’m building Ripple, a VS Code extension that generates live codebase context for AI coding agents — imports, dependents, risky files, symbol/caller awareness, and an agent-readable .ripple/context.json.

The problem I’m trying to validate is this:

Static agent files explain the repo once, but as the codebase changes, that context can become stale.

Do you feel this problem in this repo or in your own workflow?
When using Claude Code, Cursor, Copilot, Codex, or similar tools, do you ever worry that the agent is editing without understanding hidden dependencies or blast radius?

I’m looking for brutal feedback, not promotion.
For context, this is the extension I’m building:
https://marketplace.visualstudio.com/items?itemName=rippleai.ripple