Developer maps token impact of all 33 Claude Code hook events

On October 26, 2026, a developer published an article on DEV Community breaking down the 33 hook events currently listed in the Claude Code hooks reference, spanning from SessionStart to ElicitationResult, noting that the documentation was accessed on September 24, 2026. The piece observes that most people first encounter hooks through a list of ten copy-paste examples, such as running a linter after every edit, a desktop notification, or a guard on rm -rf, and then come across claims that a hook cut someone's token usage by 42% or 90%, leaving them wondering whether their own hooks help or hurt. The article stresses that both outcomes genuinely happen, and that the difference lies in which events are listened to and which JSON fields are returned. It also flags common pitfalls that cause hooks to fail silently. Along the way it gives concrete examples, including a hook that writes to stdout on UserPromptSubmit, whose text gets added to the conversation on every prompt sent. The central message for readers is that hook behavior, token overhead, and silent failures all depend on the specific event being watched and on the contents and format of what the hook returns.