[sandbox] Filter out certain non-sandbox errors (#1804)

## Summary
Users frequently complain about re-approving commands that have failed
for non-sandbox reasons. We can't diagnose with complete accuracy which
errors happened because of a sandbox failure, but we can start to
eliminate some common simple cases.

This PR captures the most common case I've seen, which is a `command not
found` error.

## Testing
- [x] Added unit tests
- [x] Ran a few cases locally

codex-rs/core/src/exec.rs

@@ -140,11 +140,7 @@ pub async fn process_exec_tool_call(
 
             let exit_code = raw_output.exit_status.code().unwrap_or(-1);
 
-            // NOTE(ragona): This is much less restrictive than the previous check. If we exec
-            // a command, and it returns anything other than success, we assume that it may have
-            // been a sandboxing error and allow the user to retry. (The user of course may choose
-            // not to retry, or in a non-interactive mode, would automatically reject the approval.)
-            if exit_code != 0 && sandbox_type != SandboxType::None {
+            if exit_code != 0 && is_likely_sandbox_denied(sandbox_type, exit_code) {
                 return Err(CodexErr::Sandbox(SandboxErr::Denied(
                     exit_code, stdout, stderr,
                 )));
@@ -223,6 +219,26 @@ fn create_linux_sandbox_command_args(
     linux_cmd
 }
 
+/// We don't have a fully deterministic way to tell if our command failed
+/// because of the sandbox - a command in the user's zshrc file might hit an
+/// error, but the command itself might fail or succeed for other reasons.
+/// For now, we conservatively check for 'command not found' (exit code 127),
+/// and can add additional cases as necessary.
+fn is_likely_sandbox_denied(sandbox_type: SandboxType, exit_code: i32) -> bool {
+    if sandbox_type == SandboxType::None {
+        return false;
+    }
+
+    // Quick rejects: well-known non-sandbox shell exit codes
+    // 127: command not found, 2: misuse of shell builtins
+    if exit_code == 127 {
+        return false;
+    }
+
+    // For all other cases, we assume the sandbox is the cause
+    true
+}
+
 #[derive(Debug)]
 pub struct RawExecToolCallOutput {
     pub exit_status: ExitStatus,

codex-rs/core/src/lib.rs

@@ -38,7 +38,7 @@ pub mod plan_tool;
 mod project_doc;
 pub mod protocol;
 mod rollout;
-mod safety;
+pub(crate) mod safety;
 pub mod seatbelt;
 pub mod shell;
 pub mod spawn;
@@ -47,3 +47,4 @@ pub mod util;
 
 pub use apply_patch::CODEX_APPLY_PATCH_ARG1;
 pub use client_common::model_supports_reasoning_summaries;
+pub use safety::get_platform_sandbox;