You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
[Docs]: document BypassSandbox and toolSummary in the PreToolUse input of run_command #1304
Runtime Version: native binary from the official installer
Summary (documentation request)
The hooks page documents the PreToolUse input of run_command as CommandLine, Cwd, WaitMsBeforeAsync (plus optional RunPersistent, RequestedTerminalID). In 1.4.4 the real input also carries:
"BypassSandbox": true when the agent asks to run the command outside the terminal sandbox (absent or false otherwise);
"toolSummary".
BypassSandbox is the only way for a hook to tell a sandboxed run from an unsandboxed one. Answering {"decision": "allow", "overwrite": {"BypassSandbox": true}} also works to send a given command outside the sandbox on purpose (e.g. a CLI that must reach a local daemon), and {"decision": "deny", ...} on BypassSandbox: true is, as far as we can tell, the only reliable way to keep allowed commands inside the sandbox.
Could you document both fields, and say whether BypassSandbox (and overwriting it) is a stable part of the hook contract?
Activity