diff --git a/agents/langchain-deepagents-code/dcode-launcher.sh b/agents/langchain-deepagents-code/dcode-launcher.sh
index bc37d25ff8c..85c87aa3ed4 100755
--- a/agents/langchain-deepagents-code/dcode-launcher.sh
+++ b/agents/langchain-deepagents-code/dcode-launcher.sh
@@ -211,16 +211,4 @@ case "${1:-}" in
   status | whoami | identity | --version | -v | -V) exec "$MANAGED_DCODE_WRAPPER" "$@" ;;
 esac
 
-# DCode's one-shot mode owns and cleans up its server lifecycle before exiting.
-# Keep that established automation path outside the interactive-session
-# supervisor; this also preserves the wrapper's exact parser diagnostics.
-_nemoclaw_dcode_args=("$@")
-for ((_nemoclaw_arg_index = 0; _nemoclaw_arg_index < ${#_nemoclaw_dcode_args[@]}; _nemoclaw_arg_index++)); do
-  _nemoclaw_arg="${_nemoclaw_dcode_args[_nemoclaw_arg_index]}"
-  case "$_nemoclaw_arg" in
-    -n | -n?* | --non-interactive | --non-interactive=*) exec "$MANAGED_DCODE_WRAPPER" "$@" ;;
-  esac
-done
-unset _nemoclaw_dcode_args _nemoclaw_arg_index _nemoclaw_arg
-
 exec /opt/venv/bin/python3 -I "$MANAGED_SESSION_SUPERVISOR" "$MANAGED_DCODE_WRAPPER" "$@"
diff --git a/docs/get-started/quickstart-langchain-deepagents-code.mdx b/docs/get-started/quickstart-langchain-deepagents-code.mdx
index bde02ee2bc7..489011e0001 100644
--- a/docs/get-started/quickstart-langchain-deepagents-code.mdx
+++ b/docs/get-started/quickstart-langchain-deepagents-code.mdx
@@ -151,7 +151,7 @@ dcode -n "Summarize this repository"
 ```
 
 The managed `dcode`, `dcode.real`, and `deepagents-code` launchers use `/opt/venv/bin/python3 -I` to run the pinned package with an isolated import path and `HOME=/sandbox`.
-For interactive sessions, each launcher supervises its own process descendants so terminal exit or disconnect terminates the associated LangGraph server tree without affecting another session.
+For interactive and headless sessions, each launcher supervises its own process descendants so completion, terminal exit, or disconnect terminates the associated LangGraph server tree without affecting another session.
 After a disconnect, the supervisor uses bounded grace periods before it kills unresponsive processes from that session.
 The supervisor runs inside the Linux OpenShell sandbox and fails closed if invoked outside Linux; the host operating system does not change this sandbox guarantee.
 They disable Deep Agents Code package update checks and the LangGraph server version check, block CLI and TUI update/install commands, and disable nested remote sandbox providers, remote async subagents, MCP commands and project auto-loading, startup commands, executable hooks, ACP mode, interpreter tool calling, shell allow-list overrides, native LangSmith tracing, and ambient OpenTelemetry exporter configuration.
diff --git a/test/deepagents-code-tui-startup-check.test.ts b/test/deepagents-code-tui-startup-check.test.ts
index bf7333ea3da..8cc33b53414 100644
--- a/test/deepagents-code-tui-startup-check.test.ts
+++ b/test/deepagents-code-tui-startup-check.test.ts
@@ -418,13 +418,15 @@ describe("Deep Agents Code TUI startup check helpers", () => {
       "10-deepagents-code-tui-startup.repeat.sanitized.log",
     );
     const processCounts = path.join(captureDir, "process-counts.txt");
-    fs.writeFileSync(processCounts, "0\n1\n0\n1\n0\n");
+    fs.writeFileSync(processCounts, "0\n1\n0\n1\n0\n1\n0\n");
 
     try {
       const result = runTuiStartupCheckHelperResult(
         [
           "sandbox_exec() { printf 'NEMOCLAW_DCODE_PROBE:deepagents\\nNEMOCLAW_DCODE_ONBOARDING:complete\\n'; }",
           "ensure_expect_available() { return 0; }",
+          'run_headless_session() { printf "PONG\\n"; }',
+          "sandbox_is_ready() { return 0; }",
           "dcode_process_count() {",
           '  value="$(sed -n "1p" "$COUNT_FILE")"',
           '  sed "1d" "$COUNT_FILE" >"$COUNT_FILE.next"',
@@ -444,6 +446,11 @@ describe("Deep Agents Code TUI startup check helpers", () => {
       const sanitizedText = fs.readFileSync(sanitizedCapture, "utf8");
       const repeatedSanitizedText = fs.readFileSync(repeatedSanitizedCapture, "utf8");
       expect(result.status).toBe(0);
+      expect(result.stdout).toContain("headless dcode request returned PONG");
+      expect(result.stdout).toContain(
+        "headless completion returned the DCode/LangGraph process count to baseline",
+      );
+      expect(result.stdout).toContain("sandbox remained Ready after headless completion");
       expect(result.stdout).toContain(
         "session 1: finite expect harness reached startup and observed exit",
       );
@@ -454,6 +461,7 @@ describe("Deep Agents Code TUI startup check helpers", () => {
       expect(sanitizedText).toContain("NEMOCLAW_TUI_READY");
       expect(sanitizedText).toContain("NEMOCLAW_TUI_EXIT_CAPTURED:130");
       expect(repeatedSanitizedText).toContain("NEMOCLAW_TUI_READY");
+      expect(repeatedSanitizedText).toContain("NEMOCLAW_TUI_EXIT_CAPTURED:130");
       expect(result.stdout).toContain(
         "session 2: DCode/LangGraph process count returned to baseline",
       );
@@ -471,6 +479,8 @@ describe("Deep Agents Code TUI startup check helpers", () => {
         [
           "sandbox_exec() { printf 'NEMOCLAW_DCODE_PROBE:deepagents\\nNEMOCLAW_DCODE_ONBOARDING:complete\\n'; }",
           "ensure_expect_available() { return 0; }",
+          'run_headless_session() { printf "PONG\\n"; }',
+          "sandbox_is_ready() { return 0; }",
           "dcode_process_count() { printf 'NEMOCLAW_DCODE_PROCESS_COUNT:0\\n'; }",
           "wait_for_dcode_process_baseline() { return 0; }",
           "run_tui_expect() {",
diff --git a/test/e2e/e2e-cloud-experimental/checks/10-deepagents-code-tui-startup.sh b/test/e2e/e2e-cloud-experimental/checks/10-deepagents-code-tui-startup.sh
index e8b5a64fe7c..a6c773b260a 100755
--- a/test/e2e/e2e-cloud-experimental/checks/10-deepagents-code-tui-startup.sh
+++ b/test/e2e/e2e-cloud-experimental/checks/10-deepagents-code-tui-startup.sh
@@ -4,10 +4,10 @@
 #
 # Case: Deep Agents Code interactive TUI startup (#5620).
 #
-# This live check runs twice against a real Deep Agents Code sandbox. It proves
-# the interactive `dcode` TUI starts in a PTY, opens and closes the `/agents`
-# modal, exits after Ctrl-C without leaking DCode/LangGraph processes, and
-# leaves only sanitized, secret-free capture artifacts.
+# This live check runs a headless request and two interactive sessions against a
+# real Deep Agents Code sandbox. It proves completed sessions do not leak
+# DCode/LangGraph processes and leaves only sanitized, secret-free capture
+# artifacts.
 #
 # shellcheck disable=SC2016
 # expect(1) Tcl: $env(...) and {...} are Tcl/sh expansion, not bash expansion.
@@ -17,6 +17,7 @@ set -euo pipefail
 SANDBOX_NAME="${SANDBOX_NAME:-${NEMOCLAW_SANDBOX_NAME:-e2e-cloud-onboard}}"
 PREFIX="10-deepagents-code-tui-startup"
 TUI_TIMEOUT="${DEEPAGENTS_TUI_TIMEOUT:-90}"
+HEADLESS_TIMEOUT="${DEEPAGENTS_HEADLESS_TIMEOUT:-120}"
 PROCESS_CLEANUP_TIMEOUT=20
 # Shell-only live check fallback for remote e2e hosts; Vitest parity coverage in
 # test/deepagents-code-tui-startup-check.test.ts pins this to secret-patterns.ts.
@@ -50,6 +51,26 @@ sandbox_exec() {
   openshell sandbox exec --name "$SANDBOX_NAME" -- bash -c "$1" 2>&1
 }
 
+run_headless_session() {
+  openshell sandbox exec --name "$SANDBOX_NAME" -- \
+    dcode -n 'Reply with exactly one word: PONG' \
+    --json --max-turns 1 --timeout "$HEADLESS_TIMEOUT" 2>&1
+}
+
+sandbox_is_ready() {
+  openshell sandbox list 2>&1 \
+    | awk -v name="$SANDBOX_NAME" '$1 == name && /Ready/ { found = 1 } END { exit(found ? 0 : 1) }'
+}
+
+wait_for_sandbox_ready() {
+  local deadline=$((SECONDS + PROCESS_CLEANUP_TIMEOUT))
+  while :; do
+    sandbox_is_ready && return 0
+    [ "$SECONDS" -ge "$deadline" ] && return 1
+    sleep 1
+  done
+}
+
 is_positive_integer() {
   [[ "$1" =~ ^[1-9][0-9]*$ ]]
 }
@@ -149,9 +170,10 @@ strip_terminal_control_sequences() {
 
 cleanup_sensitive_captures() {
   local artifact
-  for artifact in "${SENSITIVE_CAPTURE_FILES[@]}"; do
+  for artifact in "${SENSITIVE_CAPTURE_FILES[@]:-}"; do
     [ -n "$artifact" ] && rm -f -- "$artifact"
   done
+  return 0
 }
 
 make_capture_dir() {
@@ -384,6 +406,12 @@ main() {
     exit 1
   fi
 
+  if ! is_positive_integer "$HEADLESS_TIMEOUT"; then
+    fail_test "DEEPAGENTS_HEADLESS_TIMEOUT must be a positive integer"
+    printf '%s\n' "${PREFIX}: $PASSED passed, $FAILED failed"
+    exit 1
+  fi
+
   if ! command -v perl >/dev/null 2>&1; then
     fail_test "perl is required to sanitize and redact Deep Agents Code TUI captures"
     printf '%s\n' "${PREFIX}: $PASSED passed, $FAILED failed"
@@ -439,6 +467,27 @@ main() {
     exit 1
   fi
 
+  local headless_output headless_rc
+  set +e
+  headless_output="$(run_headless_session)"
+  headless_rc=$?
+  set -e
+  if [ "$headless_rc" -eq 0 ] && grep -Eiq '(^|[^A-Za-z])PONG([^A-Za-z]|$)' <<<"$headless_output"; then
+    pass "headless dcode request returned PONG"
+  else
+    fail_test "headless dcode request did not return exit-zero PONG"
+  fi
+  if wait_for_dcode_process_baseline "$baseline_process_count"; then
+    pass "headless completion returned the DCode/LangGraph process count to baseline"
+  else
+    fail_test "headless completion left the DCode/LangGraph process count above baseline after ${PROCESS_CLEANUP_TIMEOUT}s"
+  fi
+  if wait_for_sandbox_ready; then
+    pass "sandbox remained Ready after headless completion"
+  else
+    fail_test "sandbox was not Ready after headless completion"
+  fi
+
   info "Running two Deep Agents Code TUI sessions in sandbox: $SANDBOX_NAME"
   info "Capture directory: $capture_dir"
 
diff --git a/test/langchain-deepagents-code-proxy-launcher.test.ts b/test/langchain-deepagents-code-proxy-launcher.test.ts
index 29f35f361f9..1061c8c0d6e 100644
--- a/test/langchain-deepagents-code-proxy-launcher.test.ts
+++ b/test/langchain-deepagents-code-proxy-launcher.test.ts
@@ -178,17 +178,14 @@ describe("Deep Agents Code direct-exec proxy launcher", () => {
     expect(launcher.indexOf(directIdentity)).toBeLessThan(launcher.indexOf(supervisedSession));
   });
 
-  it("keeps one-shot non-interactive sessions outside the interactive supervisor", () => {
+  it("routes one-shot non-interactive sessions through the session supervisor (#6720)", () => {
     const launcher = readAgentFile("dcode-launcher.sh");
-    const nonInteractiveBypass =
-      '-n | -n?* | --non-interactive | --non-interactive=*) exec "$MANAGED_DCODE_WRAPPER" "$@" ;;';
     const supervisedSession =
       'exec /opt/venv/bin/python3 -I "$MANAGED_SESSION_SUPERVISOR" "$MANAGED_DCODE_WRAPPER" "$@"';
 
-    expect(launcher).toContain(nonInteractiveBypass);
-    expect(launcher.indexOf(nonInteractiveBypass)).toBeLessThan(
-      launcher.indexOf(supervisedSession),
-    );
+    expect(launcher).not.toMatch(/-n\s*\|[^\n]*exec "\$MANAGED_DCODE_WRAPPER"/u);
+    expect(launcher).not.toMatch(/--non-interactive[^\n]*exec "\$MANAGED_DCODE_WRAPPER"/u);
+    expect(launcher).toContain(supervisedSession);
   });
 
   it("preserves the empty-prompt failure through the installed launcher chain (#6440)", () => {