Bash sandbox auto-allow gets a new command-content check
The bash auto-allow-when-sandboxed fast path now also blocks commands matched by a new denylist check, though its data source currently returns nothing
You'll noticeTier: how much it should matter to you
2Useful: my rating, 1 to 5
3Signal: worth watching, 1 to 5
Bash SafetyArea: what it touches
ImprovementsKind: in v2.1.277,
ImprovementsSection of the release
Unclear What specifically the new command-content check looks for is not stated in the finding.
What
The fast path that automatically allows a Bash command to run without confirmation when it's sandboxed now has an extra check. In addition to the existing sandbox and read-outside-current-directory checks, it now also bails out (refuses to auto-allow) when a new function, rz(e.command), returns true for the command.
The same rz(e.command) check, along with a related fsn() lookup for substrings and commands that should always be blocked, is also consulted by a broader excluded-command check, not just the sandboxed auto-allow path.
Currently, the data source behind this check (fsn()) is a stub that returns null, so the new check has no practical effect yet.
Why
This wires in the plumbing for blocking specific commands or command substrings from being auto-allowed in a sandbox, even though the actual denylist data isn't populated yet. Until fsn() returns real data, this change does not change what gets auto-allowed in practice.
How sure we are
Something disagreesSomething we can check disagrees with this entry, or the writer said they could not settle it.
The writer flagged doubtWhat specifically the new command-content check looks for is not stated in the finding.