What exactly was the bypass that prompted the rebuild?

A command form that the hook did not parse as a git operation, combined with an operation that silently disabled the pre-push hook, let both guards be skipped at once.

How does the rebuilt Claude Code hook handle uncommon commands?

It correctly judges the usual push form and shows a confirmation dialog for unusual forms like direct URLs, other remotes, or commands prefixed with another command.

What happens if the hook's Python script crashes?

The hook now fails closed: if the Python exits abnormally or output is unexpected, a confirmation dialog appears instead of silently continuing.

AI Coding AssistantsOpen-Source AIZenn AI/ML6h agoNov 7, 2026, 19:01 JST

Claude Code main-push guard rebuilt after bypass; 55 tests pass vs 34 failures

Send on LINE

3 Key Points

What happened
After a Fable 5 review found one command form that slipped past both the Claude Code hook and the Git pre-push hook, he redesigned the hooks, and 55 test cases then passed versus 34 failures in the old version.
Why it matters
The redesign splits duties: the Claude Code hook now confirms unusual forms broadly, the pre-push hook blocks main by Git-resolved destination, and the hook fails closed on error—so a single bypass no longer defeats both layers.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, every morning.

Get these free every morning

Context & Analysis

The rebuild was prompted by a two-stage review: a general-purpose agent first, then Fable 5 checking that result. Both recommended against applying the fix as proposed, because it introduced regressions and missed command forms. The engineer's key lesson was to build regression tests before fixing a guard—comparing the old and new hooks on the same inputs—since a guard's priority is stopping what should be stopped, not just passing what should pass. He also narrowed the usual command form and made everything outside it a confirmation, rather than trying to parse every possible string. The intended adversary is mistakes and misdirection, not attackers; deliberate circumvention is left to rules and confirmation dialogs.

FAQ

What exactly was the bypass that prompted the rebuild?: A command form that the hook did not parse as a git operation, combined with an operation that silently disabled the pre-push hook, let both guards be skipped at once.
How does the rebuilt Claude Code hook handle uncommon commands?: It correctly judges the usual push form and shows a confirmation dialog for unusual forms like direct URLs, other remotes, or commands prefixed with another command.
What happens if the hook's Python script crashes?: The hook now fails closed: if the Python exits abnormally or output is unexpected, a confirmation dialog appears instead of silently continuing.

Zenn AI/MLRead Original Article

Send on LINE

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Get it free

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.