Codex forced core.fsmonitor=false on every internal Git command to
prevent repository-selected fsmonitor helpers from running. That also
disabled Git's built-in fsmonitor for status, diff, and ls-files,
turning those worktree reads into full scans in large repositories.

Probe effective core.fsmonitor only for commands that can benefit.
Preserve canonical boolean true while mapping false, unset, helper
paths, malformed values, and probe failures to false. Current Git
boolean semantics start the built-in daemon on demand and fall back to
a full scan when it cannot be queried.

Share the original five-second deadline between the probe and requested
command so the extra process cannot extend the operation's timeout. Do
not cache the decision because layered and conditional configuration can
change while Codex is running.

The git-utils wrapper now preserves built-in fsmonitor, but TUI /diff
uses the workspace command executor and continued forcing it off for
its tracked diff and untracked-file scan.

Share the config and capability output checks while leaving process
execution with each caller. Probe once per /diff and reuse the result
for tracked diff and ls-files. Keep no-index diffs disabled because
they do not inspect the index and cannot benefit from fsmonitor.

Also query core.fsmonitor without typed conversion. Git converts every
matching value before --get selects the last, so a shadowed global
helper could mask a repository-local true. Validate only the raw
effective value and cover that layered configuration.

The local Git wrapper and TUI /diff use different command runners, but
both need to reject configured fsmonitor helpers without discarding the
built-in daemon's worktree-scan acceleration.

Move config and capability interpretation into git-utils. Read the raw
effective value before normalizing it so an invalid value from a lower
configuration scope cannot poison the selected value. Accept every
boolean form Git accepts, delegating uncommon forms back to Git with an
exact-value query, and require the daemon capability before preserving
the setting.

Probe once at each worktree workflow boundary and reuse the result for
all of its commands. Keep metadata-only commands disabled without
guessing from argv, and bound each local probe and requested command
independently with the existing command timeout.

Keep probe execution in the local and workspace adapters. This preserves
their existing process and transport behavior without introducing a
generic command abstraction.

Fsmonitor detection runs before status and diff, but giving the
requested command a fresh timeout lets a stalled probe double the
existing five-second command budget.

Create one deadline at each local worktree workflow boundary and pass
it through detection and command execution. This preserves the latency
bound while still reusing one fsmonitor decision across the workflow.

Inline the probe arguments at their only call sites.

Sharing one deadline between fsmonitor detection and subsequent Git
commands changes the existing timeout behavior for diff collection.
Later ls-files and no-index diffs inherit time spent by earlier
commands.

The fsmonitor probes only query configuration and build options and do
not inspect the worktree or index. Keep them independently bounded and
preserve the existing five-second timeout for each requested Git
command.