Developer builds hook system to force AI coding agents to log their reasoning

A developer has created a three-hook system to ensure AI coding agents like Claude Code and Cursor permanently record the reasoning behind their decisions, not just the code they produce. The problem addressed is that once an agent session ends, context such as why a specific index structure or retry logic was chosen is lost forever. The solution uses an Obsidian vault synced across machines and a phone, so decision logs outlive any single session or device. Three shell scripts handle the workflow: one feeds the agent past logs at session start, another nudges it every five turns if nothing has been written, and a third blocks the session from closing if significant code changes were made without a corresponding log entry. The approach enforces documentation through technical constraints rather than relying on prompt instructions, which the developer found consistently ignored in favour of the primary coding task.