Tell HN: Claude 4.7 is ignoring stop hooks
Summary
A Hacker News thread discusses Claude 4.7's stop-hook behavior, with examples showing a stop-hook that should block a run if source files changed without tests. Participants suggest the issue may be due to stop-hook schema changes, context handling, or tool-result contextualization; some argue hooks should act as hard stops, while others discuss non-determinism and the limits of prompts, tools, and memory in LLMs. The discussion covers how to structure tests, interpret tool output safely, and design more robust harnesses to enforce discipline in AI-driven workflows.