I counted how many times my hooks had to stop Claude in one week. 448 times in 76 sessions

A user reported that their custom hooks blocked Claude Code 448 times across 76 sessions in one week, averaging about 6 blocks per session. The most frequent block, occurring 162 times, was related to Claude attempting to read files using cat and sed in bash, even after the user implemented a block for this action. This highlights an ongoing challenge in controlling AI model behavior through custom safeguards.

This report provides specific, quantified data on the frequency and types of blocks encountered when using custom hooks to control Claude Code, unlike general discussions of AI safety.