feat: add shell snapshot for shell command (#7786)

codex-rs/core/tests/suite/shell_snapshot.rs

@@ -85,14 +85,88 @@ async fn run_snapshot_command(command: &str) -> Result<SnapshotRun> {
         _ => None,
     })
     .await;
+    let mut entries = fs::read_dir(codex_home.join("shell_snapshots")).await?;
+    let snapshot_path = entries
+        .next_entry()
+        .await?
+        .map(|entry| entry.path())
+        .expect("shell snapshot created");
+    let snapshot_content = fs::read_to_string(&snapshot_path).await?;
 
-    let snapshot_arg = begin
-        .command
-        .iter()
-        .find(|arg| arg.contains("shell_snapshots"))
-        .expect("command includes shell snapshot path")
-        .to_owned();
-    let snapshot_path = PathBuf::from(&snapshot_arg);
+    let end = wait_for_event_match(&codex, |ev| match ev {
+        EventMsg::ExecCommandEnd(ev) if ev.call_id == call_id => Some(ev.clone()),
+        _ => None,
+    })
+    .await;
+
+    wait_for_event(&codex, |ev| matches!(ev, EventMsg::TaskComplete(_))).await;
+
+    Ok(SnapshotRun {
+        begin,
+        end,
+        snapshot_path,
+        snapshot_content,
+        codex_home,
+    })
+}
+
+#[allow(clippy::expect_used)]
+async fn run_shell_command_snapshot(command: &str) -> Result<SnapshotRun> {
+    let builder = test_codex().with_config(|config| {
+        config.features.enable(Feature::ShellSnapshot);
+    });
+    let harness = TestCodexHarness::with_builder(builder).await?;
+    let args = json!({
+        "command": command,
+        "timeout_ms": 1000,
+    });
+    let call_id = "shell-snapshot-command";
+    let responses = vec![
+        sse(vec![
+            ev_response_created("resp-1"),
+            ev_function_call(call_id, "shell_command", &serde_json::to_string(&args)?),
+            ev_completed("resp-1"),
+        ]),
+        sse(vec![
+            ev_response_created("resp-2"),
+            ev_assistant_message("msg-1", "done"),
+            ev_completed("resp-2"),
+        ]),
+    ];
+    mount_sse_sequence(harness.server(), responses).await;
+
+    let test = harness.test();
+    let codex = test.codex.clone();
+    let codex_home = test.home.path().to_path_buf();
+    let session_model = test.session_configured.model.clone();
+    let cwd = test.cwd_path().to_path_buf();
+
+    codex
+        .submit(Op::UserTurn {
+            items: vec![UserInput::Text {
+                text: "run shell_command with shell snapshot".into(),
+            }],
+            final_output_json_schema: None,
+            cwd,
+            approval_policy: AskForApproval::Never,
+            sandbox_policy: SandboxPolicy::DangerFullAccess,
+            model: session_model,
+            effort: None,
+            summary: ReasoningSummary::Auto,
+        })
+        .await?;
+
+    let begin = wait_for_event_match(&codex, |ev| match ev {
+        EventMsg::ExecCommandBegin(ev) if ev.call_id == call_id => Some(ev.clone()),
+        _ => None,
+    })
+    .await;
+    let mut entries = fs::read_dir(codex_home.join("shell_snapshots")).await?;
+    let snapshot_path = entries
+        .next_entry()
+        .await?
+        .map(|entry| entry.path())
+        .expect("shell snapshot created");
     let snapshot_content = fs::read_to_string(&snapshot_path).await?;
 
     let end = wait_for_event_match(&codex, |ev| match ev {
@@ -134,21 +208,9 @@ async fn linux_unified_exec_uses_shell_snapshot() -> Result<()> {
     let run = run_snapshot_command(command).await?;
     let stdout = normalize_newlines(&run.end.stdout);
 
-    let shell_path = run
-        .begin
-        .command
-        .first()
-        .expect("shell path recorded")
-        .clone();
-    assert_eq!(run.begin.command.get(1).map(String::as_str), Some("-c"));
-    assert_eq!(
-        run.begin.command.get(2).map(String::as_str),
-        Some(". \"$0\" && exec \"$@\"")
-    );
-    assert_eq!(run.begin.command.get(4), Some(&shell_path));
-    assert_eq!(run.begin.command.get(5).map(String::as_str), Some("-c"));
-    assert_eq!(run.begin.command.last(), Some(&command.to_string()));
-
+    assert_eq!(run.begin.command.get(1).map(String::as_str), Some("-lc"));
+    assert_eq!(run.begin.command.get(2).map(String::as_str), Some(command));
+    assert_eq!(run.begin.command.len(), 3);
     assert!(run.snapshot_path.starts_with(&run.codex_home));
     assert_posix_snapshot_sections(&run.snapshot_content);
     assert_eq!(run.end.exit_code, 0);
@@ -160,6 +222,93 @@ async fn linux_unified_exec_uses_shell_snapshot() -> Result<()> {
     Ok(())
 }
 
+#[cfg_attr(target_os = "windows", ignore)]
+#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+async fn linux_shell_command_uses_shell_snapshot() -> Result<()> {
+    let command = "echo shell-command-snapshot-linux";
+    let run = run_shell_command_snapshot(command).await?;
+
+    assert_eq!(run.begin.command.get(1).map(String::as_str), Some("-lc"));
+    assert_eq!(run.begin.command.get(2).map(String::as_str), Some(command));
+    assert_eq!(run.begin.command.len(), 3);
+    assert!(run.snapshot_path.starts_with(&run.codex_home));
+    assert_posix_snapshot_sections(&run.snapshot_content);
+    assert_eq!(
+        normalize_newlines(&run.end.stdout).trim(),
+        "shell-command-snapshot-linux"
+    );
+    assert_eq!(run.end.exit_code, 0);
+
+    Ok(())
+}
+
+#[cfg_attr(target_os = "windows", ignore)]
+#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+async fn shell_command_snapshot_still_intercepts_apply_patch() -> Result<()> {
+    let builder = test_codex().with_config(|config| {
+        config.features.enable(Feature::ShellSnapshot);
+        config.include_apply_patch_tool = true;
+    });
+    let harness = TestCodexHarness::with_builder(builder).await?;
+
+    let test = harness.test();
+    let codex = test.codex.clone();
+    let cwd = test.cwd_path().to_path_buf();
+    let codex_home = test.home.path().to_path_buf();
+    let target = cwd.join("snapshot-apply.txt");
+
+    let script = "apply_patch <<'EOF'\n*** Begin Patch\n*** Add File: snapshot-apply.txt\n+hello from snapshot\n*** End Patch\nEOF\n";
+    let args = json!({
+        "command": script,
+        "timeout_ms": 1_000,
+    });
+    let call_id = "shell-snapshot-apply-patch";
+    let responses = vec![
+        sse(vec![
+            ev_response_created("resp-1"),
+            ev_function_call(call_id, "shell_command", &serde_json::to_string(&args)?),
+            ev_completed("resp-1"),
+        ]),
+        sse(vec![
+            ev_response_created("resp-2"),
+            ev_assistant_message("msg-1", "done"),
+            ev_completed("resp-2"),
+        ]),
+    ];
+    mount_sse_sequence(harness.server(), responses).await;
+
+    let model = test.session_configured.model.clone();
+    codex
+        .submit(Op::UserTurn {
+            items: vec![UserInput::Text {
+                text: "apply patch via shell_command with snapshot".into(),
+            }],
+            final_output_json_schema: None,
+            cwd: cwd.clone(),
+            approval_policy: AskForApproval::Never,
+            sandbox_policy: SandboxPolicy::DangerFullAccess,
+            model,
+            effort: None,
+            summary: ReasoningSummary::Auto,
+        })
+        .await?;
+
+    wait_for_event(&codex, |ev| matches!(ev, EventMsg::TaskComplete(_))).await;
+
+    assert_eq!(fs::read_to_string(&target).await?, "hello from snapshot\n");
+
+    let mut entries = fs::read_dir(codex_home.join("shell_snapshots")).await?;
+    let snapshot_path = entries
+        .next_entry()
+        .await?
+        .map(|entry| entry.path())
+        .expect("shell snapshot created");
+    let snapshot_content = fs::read_to_string(&snapshot_path).await?;
+    assert_posix_snapshot_sections(&snapshot_content);
+
+    Ok(())
+}
+
 #[cfg_attr(not(target_os = "macos"), ignore)]
 #[cfg_attr(
     target_os = "macos",