diff --git a/tests/README.md b/tests/README.md
index f45a8a831..90cd2379d 100644
--- a/tests/README.md
+++ b/tests/README.md
@@ -61,6 +61,7 @@ All tests live directly in `integration/`; shared mechanics live in `utils/`.
 | `test_case_06_*`, `test_case_08_*` | Pass a provider or model-location override to managed Codex after configure and from fresh state | ug rejects the override before Codex starts and preserves agent-owned state |
 | `test_case_10_*`, `test_case_12_*` | Disable discovery and pass a provider or model-location override to managed Codex | ug still rejects both configured and fresh launches |
 | `test_ug_configure_managed_codex_catalog_fallback` | Configure from an injected managed response containing a GPT model absent from Codex's bundled catalog | Actionable metadata warning; conservative catalog entry for the unknown model; real Codex prompt on the valid default model |
+| `test_managed_fixture_codex_http_headers_in_managed_file` | Interactive PTY configure with injected managed `http_headers` for Codex | The specified header (`x-databricks-workspace`) lands in `model_providers.Databricks.http_headers` in `/etc/codex/managed_config.toml` with the exact admin value |
 | `test_managed_fixture_claude_mps_defaults_accompany_discovery`, `test_managed_fixture_claude_parent_schema_defaults_accompany_discovery` | Launch Claude from injected managed defaults with MPS and Unity Catalog discovery | Both generated settings files retain every admin-authored default alongside the source header; only UC Opus/Sonnet family ids gain `[1m]` |
 | `test_managed_fixture_claude_model_lifecycle`, `test_managed_fixture_codex_model_lifecycle` | Configure across no config -> static A -> static B -> MPS -> no config (stub-injected, `null` for no-config; MPS via a real provider service) | Each agent's model files reconcile to each static config (removed models pruned); switching to an MPS and a workspace with no managed config both clear ug's managed model settings so no stale list is enforced |
 | `test_ug_installed_wheel_exposes_help_and_version` | Invoke freshly installed console command | Package version matches; public help works |
@@ -72,9 +73,9 @@ All tests live directly in `integration/`; shared mechanics live in `utils/`.
 With both agents selected there are **46 live cases** (6 interactive TUI cases),
 **4 managed-workspace cases** (marker `managed`, run against a separate workspace that
 publishes a CodingAgentConfig), **1 two-workspace case** (marker `workspace_switch`),
-**36 managed-fixture cases** (marker `managed_fixture`, with only
+**37 managed-fixture cases** (marker `managed_fixture`, with only
 the CodingAgentConfig input injected), and **5 installation checks**. The 24 numbered scenarios
-cover configured and fresh state across the Claude and Codex managed-discovery matrix; twelve
+cover configured and fresh state across the Claude and Codex managed-discovery matrix; thirteen
 existing collected cases cover focused model, MCP, skills, and lifecycle shapes. Parametrization
 varies
 argument spelling or routing mode, never hides the agent/provider in the test name. Duplicate boot-only cases
diff --git a/tests/integration/README.md b/tests/integration/README.md
index 2ac6a4ddf..be2e6386c 100644
--- a/tests/integration/README.md
+++ b/tests/integration/README.md
@@ -178,10 +178,10 @@ checks** with both agents. A separate **4 managed-workspace cases** (one per age
 re-configure, and a cache-TTL journey; marker `managed`) run against a workspace that publishes a
 CodingAgentConfig; see "Managed-workspace journeys" below. One **`workspace_switch` case**
 uses two real workspaces and checks skills MCP cleanup and a completed Claude task.
-A further **36 `managed_fixture`
+A further **37 `managed_fixture`
 cases** use `UCODE_MANAGED_CONFIG_STUB`. Twenty-four explicit configured/fresh Claude and Codex
 discovery and source-override journeys fetch the published config once per agent, replace that
-agent's static source with its dedicated MPS, and reuse the result. Twelve existing collected cases
+agent's static source with its dedicated MPS, and reuse the result. Thirteen existing collected cases
 cover focused model, MCP, skills, and lifecycle shapes, including per-agent model reconciliation
 and managed skill cleanup. The two Claude default-model cases launch with injected MPS and Unity
 Catalog sources and verify both generated settings files retain all admin-authored family defaults.
diff --git a/tests/integration/test_ug_configure_managed_mcp.py b/tests/integration/test_ug_configure_managed_mcp.py
index 93268805f..b3d66f572 100644
--- a/tests/integration/test_ug_configure_managed_mcp.py
+++ b/tests/integration/test_ug_configure_managed_mcp.py
@@ -93,3 +93,46 @@ def test_managed_fixture_codex_mcp_written_to_managed_file(live_session, workspa
     personal = tomllib.loads(personal_path.read_text()) if personal_path.exists() else {}
     personal_servers = personal.get("mcp_servers") or {}
     assert not any("github" in name.lower() for name in personal_servers), personal_servers
+
+
+@pytest.mark.managed_fixture
+@pytest.mark.codex
+def test_managed_fixture_codex_http_headers_in_managed_file(live_session, workspace, tmp_path):
+    """Scenario: an interactive configure with managed http_headers writes them into the OS-managed file.
+
+    Scope: managed-config content assertion in the same family as the MCP/skills/models
+    managed_fixture tests — verifies the http_headers field flows from the injected
+    CodingAgentConfig through `ug configure codex` into the OS-managed
+    /etc/codex/managed_config.toml under [model_providers.Databricks.http_headers].
+
+    Expected: after an interactive PTY configure, /etc/codex/managed_config.toml contains
+    model_providers.Databricks.http_headers with the admin-specified header
+    x-databricks-workspace = "eng-ml-inference", exactly as the injected managed config
+    dictates.
+    """
+    session = live_session
+    managed_header_key = "x-databricks-workspace"
+    managed_header_value = "eng-ml-inference"
+    config = build_coding_agent_config(
+        "CODING_AGENT_CODEX",
+        build_codex_agent_config(
+            models=[CODEX_MODEL],
+            http_headers={managed_header_key: managed_header_value},
+        ),
+    )
+    set_managed_config_stub(session, tmp_path, config)
+    command = [str(session.binary), "configure", "--workspace", workspace, "--skip-upgrade"]
+    with ConfigureTerminal(session, "codex", command, "managed-http-headers-codex") as configure:
+        configure.finish(timeout=300)
+
+    managed = tomllib.loads(session.run(CODEX_MANAGED_CONFIG_PATH, binary="cat", timeout=30).stdout)
+    provider = (managed.get("model_providers") or {}).get("Databricks") or {}
+    headers = provider.get("http_headers") or {}
+    assert managed_header_key in headers, (
+        f"Expected header {managed_header_key!r} in model_providers.Databricks.http_headers; "
+        f"got: {headers!r}\nFull managed config: {managed!r}"
+    )
+    assert headers[managed_header_key] == managed_header_value, (
+        f"Expected {managed_header_key!r} = {managed_header_value!r}, "
+        f"got {headers[managed_header_key]!r}"
+    )
diff --git a/tests/integration/utils/managed.py b/tests/integration/utils/managed.py
index 423ebcf80..fbeb86630 100644
--- a/tests/integration/utils/managed.py
+++ b/tests/integration/utils/managed.py
@@ -115,13 +115,20 @@ def build_claude_agent_config(
     return {"agent": "CODING_AGENT_CLAUDE_CODE", "config": config}
 
 
-def build_codex_agent_config(*, models: list[str], smart_routing: bool = False) -> dict:
+def build_codex_agent_config(
+    *,
+    models: list[str],
+    smart_routing: bool = False,
+    http_headers: dict[str, str] | None = None,
+) -> dict:
     config = {
         "models": {"model_services": models},
         "default_models": {"default_model": models[0]},
     }
     if smart_routing:
         config["smart_routing"] = {"enabled": True}
+    if http_headers is not None:
+        config["http_headers"] = http_headers
     return {"agent": "CODING_AGENT_CODEX", "config": config}