Keeping context and decisions consistent across parallel AI agents

You start the morning with four Claude Code agents running, each in its own git worktree, each on a separate task. By mid-afternoon something is off. One agent has re-implemented a helper another alre...