Stop your AI agent from ignoring your architecture

An AI agent makes architectural decisions constantly. Add a dependency, change a build script, restructure a config file. Each choice is reasonable in isolation. None of them get written down. This is the knowledge management version of technical debt. Six months later, someone asks why the project uses rehype-highlight instead of Shiki. The answer is in a conversation that no longer exists. The decision was sound. The reasoning is gone. A hook-based gate can close this gap. This implementation uses Claude Code hooks ( source code ), but the pattern (detect, gate, review, unlock, reset) applies to any agent system that exposes lifecycle events. Claude Code hooks are shell scripts that run at specific points in the agent's lifecycle. UserPromptSubmit fires when the user sends a message. PreToolUse fires before the agent calls a tool (Edit, Write, Bash). PostToolUse fires after a tool call completes. Stop fires when the agent finishes its turn. Each hook receives JSON on stdin describing