[hardy30894]

Why

Claude Code evaluates permission rules against the input a PreToolUse hook returns, not the input the model sent (hooks reference, "PreToolUse decision control"). secondwind hook rewrote every Bash command to secondwind exec -c '<command>' and answered "permissionDecision": "allow". Two consequences for anyone who ran secondwind setup --hook:

• Bash allow and deny rules no longer matched: a rule like Bash(rm:*) does not match secondwind exec -c 'rm -rf build'.
• allow skipped the permission prompt for every Bash command.

There is no hook decision that rewrites the input and leaves the decision to the normal flow: ask prompts on every command, defer needs -p and ignores the rewrite, and omitting the decision is undocumented. The hook cannot be fixed, so it is retired. The Claude Code mod in integrations/claude-code compresses Bash output after the command has run and never touches permissions.

What

• secondwind hook reads the event and prints nothing, so Claude Code runs the command as written under its own rules. Settings that still name the hook keep working.
• secondwind setup --hook installs nothing and says so; plain secondwind setup removes a hook an earlier version installed, leaving any other hooks in place.
• secondwind check warns when the retired hook is still installed.
• Help text, the dashboard and the README no longer point to the hook.

secondwind exec is unchanged.

Tests

The hook answers a Bash event with nothing and exit 0; setup --hook installs nothing; setup removes a stale hook. cargo test -p secondwind: 24 passed. Clippy clean.