core: allow excluding tool namespaces from code mode (#26320)

## Why

Research and training setups need to control which tool namespaces
appear inside code mode's nested `tools` surface without disabling those
tools entirely. This makes it possible to train against a deliberately
reduced nested-tool setup while preserving the normal direct and
deferred tool paths.

## What

- Extend `features.code_mode` to accept structured configuration while
preserving the existing boolean syntax.
- Add an exact `excluded_tool_namespaces` list under
`[features.code_mode]`:

  ```toml
  [features.code_mode]
  enabled = true
  excluded_tool_namespaces = ["mcp__codex_apps", "multi_agent_v1"]
  ```

- Filter matching canonical `ToolName` namespaces when constructing code
mode's nested router and code-mode-specific direct tool descriptions.
- Keep excluded tools registered, directly exposed in mixed code mode,
and discoverable through top-level `tool_search` when otherwise
eligible.
- Derive deferred nested-tool guidance after namespace filtering so the
`exec` description does not advertise excluded-only deferred tools.
- Preserve the boolean/table representation when materializing config
locks and update the generated config schema.

## Testing

- `just test -p codex-features`
- `just test -p codex-config`
- `just test -p codex-core load_config_resolves_code_mode_config`
- `just test -p codex-core
lock_contains_prompts_and_materializes_features`
- `just test -p codex-core
excluded_deferred_namespaces_do_not_enable_nested_tool_guidance`
- `just test -p codex-core
code_mode_excludes_configured_nested_tool_namespaces`
- `cargo check -p codex-thread-manager-sample`
This commit is contained in:
sayan-oai
2026-06-04 11:40:18 -07:00
committed by GitHub
Unverified
parent 9e41f8ddbe
commit 8b1238856b
11 changed files with 286 additions and 16 deletions
+9
View File
@@ -25,6 +25,15 @@ pub fn features_schema(schema_gen: &mut SchemaGenerator) -> Schema {
if feature.id == codex_features::Feature::Artifact {
continue;
}
if feature.id == codex_features::Feature::CodeMode {
validation.properties.insert(
feature.key.to_string(),
schema_gen.subschema_for::<codex_features::FeatureToml<
codex_features::CodeModeConfigToml,
>>(),
);
continue;
}
if feature.id == codex_features::Feature::MultiAgentV2 {
validation.properties.insert(
feature.key.to_string(),
+28 -2
View File
@@ -355,6 +355,22 @@
},
"type": "object"
},
"CodeModeConfigToml": {
"additionalProperties": false,
"properties": {
"enabled": {
"type": "boolean"
},
"excluded_tool_namespaces": {
"description": "Exact tool namespaces to omit from the code-mode nested tool surface.",
"items": {
"type": "string"
},
"type": "array"
}
},
"type": "object"
},
"ConfigProfile": {
"additionalProperties": false,
"description": "Collection of common configuration options that a user can define as a unit in `config.toml`.",
@@ -410,7 +426,7 @@
"type": "boolean"
},
"code_mode": {
"type": "boolean"
"$ref": "#/definitions/FeatureToml_for_CodeModeConfigToml"
},
"code_mode_only": {
"type": "boolean"
@@ -830,6 +846,16 @@
}
]
},
"FeatureToml_for_CodeModeConfigToml": {
"anyOf": [
{
"type": "boolean"
},
{
"$ref": "#/definitions/CodeModeConfigToml"
}
]
},
"FeatureToml_for_MultiAgentV2ConfigToml": {
"anyOf": [
{
@@ -4533,7 +4559,7 @@
"type": "boolean"
},
"code_mode": {
"type": "boolean"
"$ref": "#/definitions/FeatureToml_for_CodeModeConfigToml"
},
"code_mode_only": {
"type": "boolean"
+26
View File
@@ -476,6 +476,32 @@ async fn load_config_resolves_experimental_request_user_input_enabled() -> std::
Ok(())
}
#[tokio::test]
async fn load_config_resolves_code_mode_config() -> std::io::Result<()> {
let codex_home = tempdir()?;
let config_toml: ConfigToml = toml::from_str(
r#"
[features.code_mode]
enabled = true
excluded_tool_namespaces = ["mcp__codex_apps", "multi_agent_v1"]
"#,
)
.expect("TOML deserialization should succeed");
let config = Config::load_from_base_config_with_overrides(
config_toml,
ConfigOverrides::default(),
codex_home.abs(),
)
.await?;
assert_eq!(
config.code_mode.excluded_tool_namespaces,
vec!["mcp__codex_apps".to_string(), "multi_agent_v1".to_string()]
);
assert!(config.features.enabled(Feature::CodeMode));
Ok(())
}
#[test]
fn rejects_provider_auth_with_env_key() {
let err = toml::from_str::<ConfigToml>(
+29
View File
@@ -58,6 +58,7 @@ use codex_core_plugins::PluginsConfigInput;
use codex_exec_server::ExecutorFileSystem;
use codex_exec_server::LOCAL_FS;
use codex_features::AppsMcpPathOverrideConfigToml;
use codex_features::CodeModeConfigToml;
use codex_features::Feature;
use codex_features::FeatureConfigSource;
use codex_features::FeatureOverrides;
@@ -977,6 +978,9 @@ pub struct Config {
/// Whether to register the experimental request_user_input tool.
pub experimental_request_user_input_enabled: bool,
/// Configuration for the experimental code-mode tool surface.
pub code_mode: CodeModeConfig,
/// If set to `true`, used only the experimental unified exec tool.
pub use_experimental_unified_exec_tool: bool,
@@ -1029,6 +1033,11 @@ pub struct Config {
pub otel: codex_config::types::OtelConfig,
}
#[derive(Debug, Clone, Default, PartialEq, Eq, Serialize)]
pub struct CodeModeConfig {
pub excluded_tool_namespaces: Vec<String>,
}
#[derive(Debug, Clone, PartialEq, Eq, Serialize)]
pub struct MultiAgentV2Config {
pub max_concurrent_threads_per_session: usize,
@@ -2290,6 +2299,17 @@ fn resolve_experimental_request_user_input_enabled(config_toml: &ConfigToml) ->
.is_none_or(|config| config.enabled)
}
fn resolve_code_mode_config(config_toml: &ConfigToml) -> CodeModeConfig {
let base = code_mode_toml_config(config_toml.features.as_ref());
CodeModeConfig {
excluded_tool_namespaces: base
.and_then(|config| config.excluded_tool_namespaces.as_ref())
.cloned()
.unwrap_or_default(),
}
}
fn resolve_multi_agent_v2_config(config_toml: &ConfigToml) -> MultiAgentV2Config {
let base = multi_agent_v2_toml_config(config_toml.features.as_ref());
let default = MultiAgentV2Config::default();
@@ -2372,6 +2392,13 @@ fn resolve_optional_prompt_text(
}
}
fn code_mode_toml_config(features: Option<&FeaturesToml>) -> Option<&CodeModeConfigToml> {
match features?.code_mode.as_ref()? {
FeatureToml::Enabled(_) => None,
FeatureToml::Config(config) => Some(config),
}
}
fn multi_agent_v2_toml_config(features: Option<&FeaturesToml>) -> Option<&MultiAgentV2ConfigToml> {
match features?.multi_agent_v2.as_ref()? {
FeatureToml::Enabled(_) => None,
@@ -3013,6 +3040,7 @@ impl Config {
let web_search_config = resolve_web_search_config(&cfg);
let experimental_request_user_input_enabled =
resolve_experimental_request_user_input_enabled(&cfg);
let code_mode = resolve_code_mode_config(&cfg);
let multi_agent_v2 = resolve_multi_agent_v2_config(&cfg);
let apps_mcp_path_override = if features.enabled(Feature::AppsMcpPathOverride) {
let base = apps_mcp_path_override_toml_config(cfg.features.as_ref());
@@ -3554,6 +3582,7 @@ impl Config {
web_search_mode: constrained_web_search_mode.value,
web_search_config,
experimental_request_user_input_enabled,
code_mode,
use_experimental_unified_exec_tool,
background_terminal_max_timeout,
ghost_snapshot,
+8
View File
@@ -251,6 +251,14 @@ mod tests {
spec.key
);
}
assert_eq!(
features.code_mode,
Some(FeatureToml::Enabled(
sc.original_config_do_not_use
.features
.enabled(Feature::CodeMode)
))
);
let multi_agent_v2 = features
.multi_agent_v2
+32 -13
View File
@@ -207,7 +207,12 @@ fn build_model_visible_specs_and_registry(
if exposure.is_direct() && !is_hidden_by_code_mode_only(turn_context, &tool_name, exposure)
{
let spec = runtime.spec();
specs.push(spec_for_model_request(turn_context, exposure, spec));
specs.push(spec_for_model_request(
turn_context,
exposure,
&tool_name,
spec,
));
}
}
specs.extend(hosted_specs);
@@ -226,12 +231,14 @@ fn build_model_visible_specs_and_registry(
fn spec_for_model_request(
turn_context: &TurnContext,
exposure: ToolExposure,
tool_name: &ToolName,
spec: ToolSpec,
) -> ToolSpec {
if matches!(
turn_context.tool_mode,
ToolMode::CodeMode | ToolMode::CodeModeOnly
) && exposure != ToolExposure::DirectModelOnly
&& !is_excluded_from_code_mode(turn_context, tool_name)
&& codex_code_mode::is_code_mode_nested_tool(spec.name())
{
codex_tools::augment_tool_spec_for_code_mode(spec)
@@ -405,10 +412,19 @@ fn is_hidden_by_code_mode_only(
))
}
fn is_excluded_from_code_mode(turn_context: &TurnContext, tool_name: &ToolName) -> bool {
tool_name.namespace.as_ref().is_some_and(|namespace| {
turn_context
.config
.code_mode
.excluded_tool_namespaces
.contains(namespace)
})
}
fn build_code_mode_executors(
turn_context: &TurnContext,
executors: &[Arc<dyn CoreToolRuntime>],
deferred_tools_available: bool,
) -> Vec<Arc<dyn CoreToolRuntime>> {
if !matches!(
turn_context.tool_mode,
@@ -419,6 +435,8 @@ fn build_code_mode_executors(
let mut code_mode_nested_tool_specs = Vec::new();
let mut exec_prompt_tool_specs = Vec::new();
let mut deferred_tools_available = false;
let deferred_tools_guidance_enabled = search_tool_enabled(turn_context);
for executor in executors {
let exposure = executor.exposure();
if exposure == ToolExposure::DirectModelOnly {
@@ -428,9 +446,19 @@ fn build_code_mode_executors(
if exposure == ToolExposure::Hidden {
continue;
}
if is_excluded_from_code_mode(turn_context, &executor.tool_name()) {
continue;
}
let spec = executor.spec();
if exposure != ToolExposure::Deferred {
if exposure == ToolExposure::Deferred {
// Only show deferred-tool guidance when supported and an included spec is usable by code mode.
deferred_tools_available |= deferred_tools_guidance_enabled
&& !collect_code_mode_exec_prompt_tool_definitions(std::iter::once(&spec))
.is_empty();
} else {
exec_prompt_tool_specs.push(spec.clone());
}
code_mode_nested_tool_specs.push(spec);
@@ -839,16 +867,7 @@ fn prepend_code_mode_executors(
planned_tools: &mut PlannedTools,
) {
let turn_context = context.turn_context;
let deferred_tools_available = search_tool_enabled(turn_context)
&& planned_tools
.runtimes()
.iter()
.any(|executor| executor.exposure() == ToolExposure::Deferred);
let code_mode_executors = build_code_mode_executors(
turn_context,
planned_tools.runtimes(),
deferred_tools_available,
);
let code_mode_executors = build_code_mode_executors(turn_context, planned_tools.runtimes());
planned_tools.runtimes.splice(0..0, code_mode_executors);
}
@@ -938,6 +938,42 @@ async fn code_mode_only_exposes_code_executor_and_hides_nested_tools() {
);
}
#[tokio::test]
async fn excluded_deferred_namespaces_do_not_enable_nested_tool_guidance() {
let plan = probe_with(
|turn| {
set_features(turn, &[Feature::CodeMode, Feature::CodeModeOnly]);
set_feature(turn, Feature::Collab, /*enabled*/ false);
turn.model_info.supports_search_tool = true;
update_config(turn, |config| {
config.code_mode.excluded_tool_namespaces = vec!["excluded".to_string()];
});
},
ToolPlanInputs {
dynamic_tools: vec![dynamic_tool(
Some("excluded"),
"lookup",
/*defer_loading*/ true,
)],
..ToolPlanInputs::default()
},
)
.await;
let ToolSpec::Freeform(exec) = plan.visible_spec(codex_code_mode::PUBLIC_TOOL_NAME) else {
panic!("expected code mode exec tool");
};
assert!(
!exec
.description
.contains("Some deferred nested tools may be omitted")
);
plan.assert_registered_contains(&[
&ToolName::namespaced("excluded", "lookup").to_string(),
"tool_search",
]);
}
#[tokio::test]
async fn multi_agent_feature_selects_one_agent_tool_family() {
let v1 = probe(|turn| {
+87
View File
@@ -3165,6 +3165,93 @@ text(
Ok(())
}
#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
async fn code_mode_excludes_configured_nested_tool_namespaces() -> Result<()> {
skip_if_no_network!(Ok(()));
let server = responses::start_mock_server().await;
let mut builder = test_codex().with_config(|config| {
let _ = config.features.enable(Feature::CodeMode);
config.code_mode.excluded_tool_namespaces = vec!["excluded".to_string()];
});
let base_test = builder.build(&server).await?;
let new_thread = base_test
.thread_manager
.start_thread_with_tools(
base_test.config.clone(),
vec![DynamicToolSpec {
namespace: Some("excluded".to_string()),
name: "lookup".to_string(),
description: "An excluded dynamic tool.".to_string(),
input_schema: serde_json::json!({
"type": "object",
"properties": {},
"additionalProperties": false,
}),
defer_loading: false,
}],
)
.await?;
let mut test = base_test;
test.codex = new_thread.thread;
test.session_configured = new_thread.session_configured;
let first_mock = responses::mount_sse_once(
&server,
sse(vec![
ev_response_created("resp-1"),
ev_custom_tool_call(
"call-1",
"exec",
r#"
text(JSON.stringify({
excludedType: typeof tools.excluded__lookup,
excludedMetadata: ALL_TOOLS.some(({ name }) => name === "excluded__lookup"),
allowedType: typeof tools.update_plan,
allowedMetadata: ALL_TOOLS.some(({ name }) => name === "update_plan"),
}));
"#,
),
ev_completed("resp-1"),
]),
)
.await;
let second_mock = responses::mount_sse_once(
&server,
sse(vec![
ev_assistant_message("msg-1", "done"),
ev_completed("resp-2"),
]),
)
.await;
test.submit_turn("use exec to inspect nested tool namespaces")
.await?;
assert!(
tool_names(&first_mock.single_request().body_json()).contains(&"excluded".to_string()),
"excluded namespace should remain directly exposed in mixed code mode"
);
let request = second_mock.single_request();
let (output, success) = custom_tool_output_body_and_success(&request, "call-1");
assert_ne!(
success,
Some(false),
"exec configured namespace exclusion failed unexpectedly: {output}"
);
assert_eq!(
serde_json::from_str::<Value>(&output)?,
serde_json::json!({
"excludedType": "undefined",
"excludedMetadata": false,
"allowedType": "function",
"allowedMetadata": true,
})
);
Ok(())
}
#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
async fn code_mode_can_print_content_only_mcp_tool_result_fields() -> Result<()> {
skip_if_no_network!(Ok(()));
+20
View File
@@ -4,6 +4,26 @@ use serde::Deserialize;
use serde::Serialize;
use std::collections::BTreeMap;
#[derive(Serialize, Deserialize, Debug, Clone, Default, PartialEq, Eq, JsonSchema)]
#[serde(deny_unknown_fields)]
pub struct CodeModeConfigToml {
#[serde(skip_serializing_if = "Option::is_none")]
pub enabled: Option<bool>,
/// Exact tool namespaces to omit from the code-mode nested tool surface.
#[serde(skip_serializing_if = "Option::is_none")]
pub excluded_tool_namespaces: Option<Vec<String>>,
}
impl FeatureConfig for CodeModeConfigToml {
fn enabled(&self) -> Option<bool> {
self.enabled
}
fn set_enabled(&mut self, enabled: bool) {
self.enabled = Some(enabled);
}
}
#[derive(Serialize, Deserialize, Debug, Clone, Default, PartialEq, Eq, JsonSchema)]
#[serde(deny_unknown_fields)]
pub struct MultiAgentV2ConfigToml {
+10 -1
View File
@@ -17,6 +17,7 @@ use toml::Table;
mod feature_configs;
mod legacy;
pub use feature_configs::AppsMcpPathOverrideConfigToml;
pub use feature_configs::CodeModeConfigToml;
pub use feature_configs::MultiAgentV2ConfigToml;
pub use feature_configs::NetworkProxyConfigToml;
pub use feature_configs::NetworkProxyDomainPermissionToml;
@@ -600,6 +601,8 @@ pub fn is_known_feature_key(key: &str) -> bool {
/// Deserializable features table for TOML.
#[derive(Serialize, Deserialize, Debug, Clone, Default, PartialEq, JsonSchema)]
pub struct FeaturesToml {
#[serde(default, skip_serializing_if = "Option::is_none")]
pub code_mode: Option<FeatureToml<CodeModeConfigToml>>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub multi_agent_v2: Option<FeatureToml<MultiAgentV2ConfigToml>>,
#[serde(default, skip_serializing_if = "Option::is_none")]
@@ -620,6 +623,9 @@ impl Features {
impl FeaturesToml {
pub fn entries(&self) -> BTreeMap<String, bool> {
let mut entries = self.entries.clone();
if let Some(enabled) = self.code_mode.as_ref().and_then(FeatureToml::enabled) {
entries.insert(Feature::CodeMode.key().to_string(), enabled);
}
if let Some(enabled) = self.multi_agent_v2.as_ref().and_then(FeatureToml::enabled) {
entries.insert(Feature::MultiAgentV2.key().to_string(), enabled);
}
@@ -638,6 +644,7 @@ impl FeaturesToml {
pub fn materialize_resolved_enabled(&mut self, features: &Features) {
let Self {
code_mode,
multi_agent_v2,
apps_mcp_path_override,
network_proxy,
@@ -648,7 +655,9 @@ impl FeaturesToml {
}
for spec in FEATURES {
let enabled = features.enabled(spec.id);
if spec.id == Feature::MultiAgentV2 {
if spec.id == Feature::CodeMode {
materialize_resolved_feature_enabled(code_mode, enabled);
} else if spec.id == Feature::MultiAgentV2 {
materialize_resolved_feature_enabled(multi_agent_v2, enabled);
} else if spec.id == Feature::AppsMcpPathOverride {
materialize_resolved_feature_enabled(apps_mcp_path_override, enabled);
@@ -266,6 +266,7 @@ fn new_config(model: Option<String>, arg0_paths: Arg0DispatchPaths) -> anyhow::R
web_search_mode: Constrained::allow_any(WebSearchMode::Disabled),
web_search_config: None,
experimental_request_user_input_enabled: true,
code_mode: Default::default(),
use_experimental_unified_exec_tool: false,
background_terminal_max_timeout: 300_000,
ghost_snapshot: GhostSnapshotConfig::default(),