diff --git a/codex-rs/app-server/tests/suite/v2/turn_start.rs b/codex-rs/app-server/tests/suite/v2/turn_start.rs index dd211a5d8..622e4284e 100644 --- a/codex-rs/app-server/tests/suite/v2/turn_start.rs +++ b/codex-rs/app-server/tests/suite/v2/turn_start.rs @@ -100,6 +100,7 @@ const DEFAULT_READ_TIMEOUT: std::time::Duration = std::time::Duration::from_secs const DEFAULT_READ_TIMEOUT: std::time::Duration = std::time::Duration::from_secs(10); const TEST_ORIGINATOR: &str = "codex_vscode"; const LOCAL_PRAGMATIC_TEMPLATE: &str = "You are a deeply pragmatic, effective software engineer."; +const MULTI_AGENT_V2_NAMESPACE: &str = "collaboration"; const INVALID_REQUEST_ERROR_CODE: i64 = -32600; const TINY_PNG_BYTES: &[u8] = &[ 137, 80, 78, 71, 13, 10, 26, 10, 0, 0, 0, 13, 73, 72, 68, 82, 0, 0, 0, 1, 0, 0, 0, 1, 8, 6, 0, @@ -3650,7 +3651,12 @@ async fn direct_input_to_multi_agent_v2_subagent_is_rejected() -> Result<()> { |req: &wiremock::Request| body_contains(req, PARENT_PROMPT), responses::sse(vec![ responses::ev_response_created("resp-parent-1"), - responses::ev_function_call(SPAWN_CALL_ID, "spawn_agent", &spawn_args), + responses::ev_function_call_with_namespace( + SPAWN_CALL_ID, + MULTI_AGENT_V2_NAMESPACE, + "spawn_agent", + &spawn_args, + ), responses::ev_completed("resp-parent-1"), ]), ) diff --git a/codex-rs/core/src/config/mod.rs b/codex-rs/core/src/config/mod.rs index 953d0af1a..81d2870c1 100644 --- a/codex-rs/core/src/config/mod.rs +++ b/codex-rs/core/src/config/mod.rs @@ -246,7 +246,8 @@ Payload: ``` You may also see them addressed as to=/root/..., which indicates your identity is /root/... "#; -const DEFAULT_MULTI_AGENT_V2_SHARED_USAGE_HINT_TEXT: &str = r#"Note that collaboration tools cannot be called from inside `functions.exec`. Call `spawn_agent`, `send_message`, `followup_task`, `wait_agent`, `interrupt_agent`, and `list_agents` only as direct tool calls using the recipient shown in their tool definitions, such as `to=functions.spawn_agent` without a configured namespace or `to=functions.agents.spawn_agent` with `tool_namespace = "agents"`, since they are intentionally absent from the `functions.exec` `tools.*` namespace. Available tools in `functions.exec` are explicitly described with a `tools` namespace in the developer message. +const DEFAULT_MULTI_AGENT_V2_TOOL_NAMESPACE: &str = "collaboration"; +const DEFAULT_MULTI_AGENT_V2_SHARED_USAGE_HINT_TEXT: &str = r#"Note that collaboration tools cannot be called from inside `functions.exec`. Call `spawn_agent`, `send_message`, `followup_task`, `wait_agent`, `interrupt_agent`, and `list_agents` only as direct tool calls using the recipient shown in their tool definitions, such as `to=functions.collaboration.spawn_agent`, since they are intentionally absent from the `functions.exec` `tools.*` namespace. Available tools in `functions.exec` are explicitly described with a `tools` namespace in the developer message. All agents share the same directory. In detail: - All agents have access to the same container and filesystem as you. @@ -1157,7 +1158,7 @@ impl MultiAgentV2Config { DEFAULT_MULTI_AGENT_V2_SUBAGENT_USAGE_HINT_TEXT, max_concurrent_threads_per_session, )), - tool_namespace: None, + tool_namespace: Some(DEFAULT_MULTI_AGENT_V2_TOOL_NAMESPACE.to_string()), hide_spawn_agent_metadata: true, non_code_mode_only: true, } diff --git a/codex-rs/core/src/tools/spec_plan_tests.rs b/codex-rs/core/src/tools/spec_plan_tests.rs index 5d70fb8f1..3fe33b141 100644 --- a/codex-rs/core/src/tools/spec_plan_tests.rs +++ b/codex-rs/core/src/tools/spec_plan_tests.rs @@ -39,6 +39,8 @@ use crate::tools::router::ToolRouterParams; use crate::tools::router::ToolSuggestCandidates; use crate::tools::router::ToolSuggestPresentation; +const MULTI_AGENT_V2_NAMESPACE: &str = "collaboration"; + #[derive(Default)] struct ToolPlanInputs { mcp_tools: Option>, @@ -1151,19 +1153,48 @@ async fn multi_agent_feature_selects_one_agent_tool_family() { }); }) .await; - v2.assert_visible_contains(&[ + v2.assert_visible_contains(&[MULTI_AGENT_V2_NAMESPACE]); + v2.assert_visible_lacks(&[ "spawn_agent", "send_message", "followup_task", "wait_agent", "interrupt_agent", "list_agents", + "send_input", + "resume_agent", + "assign_task", + "close_agent", ]); - v2.assert_visible_lacks(&["send_input", "resume_agent", "assign_task", "close_agent"]); - let spawn_agent_description = match v2.visible_spec("spawn_agent") { - ToolSpec::Function(tool) => tool.description.as_str(), - other => panic!("expected spawn_agent function spec, got {other:?}"), + for tool_name in [ + "spawn_agent", + "send_message", + "followup_task", + "wait_agent", + "interrupt_agent", + "list_agents", + ] { + assert!( + v2.namespace_function_names(MULTI_AGENT_V2_NAMESPACE) + .iter() + .any(|name| name == tool_name), + "expected {tool_name} in {MULTI_AGENT_V2_NAMESPACE} namespace" + ); + } + let ToolSpec::Namespace(namespace) = v2.visible_spec(MULTI_AGENT_V2_NAMESPACE) else { + panic!("expected {MULTI_AGENT_V2_NAMESPACE} namespace"); }; + let Some(ResponsesApiNamespaceTool::Function(spawn_agent)) = + namespace.tools.iter().find(|tool| { + matches!( + tool, + ResponsesApiNamespaceTool::Function(tool) if tool.name == "spawn_agent" + ) + }) + else { + panic!("expected spawn_agent in {MULTI_AGENT_V2_NAMESPACE} namespace"); + }; + let spawn_agent_description = spawn_agent.description.as_str(); assert!(!spawn_agent_description.contains("max_concurrent_threads_per_session")); assert!(spawn_agent_description.contains( "Note that passing `fork_turns=\"none\"` will not pass any surrounding context to the spawned subagent" @@ -1183,9 +1214,11 @@ async fn multi_agent_feature_selects_one_agent_tool_family() { }); }) .await; - direct_model_only.assert_visible_contains(&["spawn_agent", "send_message", "wait_agent"]); + direct_model_only.assert_visible_contains(&[MULTI_AGENT_V2_NAMESPACE]); + direct_model_only.assert_visible_lacks(&["spawn_agent", "send_message", "wait_agent"]); assert_eq!( - direct_model_only.exposure("spawn_agent"), + direct_model_only + .exposure(&ToolName::namespaced(MULTI_AGENT_V2_NAMESPACE, "spawn_agent").to_string()), ToolExposure::DirectModelOnly ); } @@ -1196,9 +1229,17 @@ async fn multi_agent_v2_message_schemas_are_encrypted() { set_feature(turn, Feature::MultiAgentV2, /*enabled*/ true); }) .await; + let ToolSpec::Namespace(namespace) = plan.visible_spec(MULTI_AGENT_V2_NAMESPACE) else { + panic!("expected {MULTI_AGENT_V2_NAMESPACE} namespace"); + }; for tool_name in ["spawn_agent", "send_message", "followup_task"] { - let ToolSpec::Function(tool) = plan.visible_spec(tool_name) else { - panic!("expected {tool_name} function spec"); + let Some(ResponsesApiNamespaceTool::Function(tool)) = namespace.tools.iter().find(|tool| { + matches!( + tool, + ResponsesApiNamespaceTool::Function(tool) if tool.name == tool_name + ) + }) else { + panic!("expected {tool_name} in {MULTI_AGENT_V2_NAMESPACE} namespace"); }; let properties = tool .parameters @@ -1464,12 +1505,7 @@ async fn hosted_tools_follow_provider_auth_model_and_config_gates() { codex_code_mode::WAIT_TOOL_NAME, "request_user_input", // Multi-agent v2 tools. - "spawn_agent", - "send_message", - "followup_task", - "wait_agent", - "interrupt_agent", - "list_agents", + MULTI_AGENT_V2_NAMESPACE, // Hosted Responses tools. "web_search", "image_generation", diff --git a/codex-rs/core/tests/suite/agent_execution.rs b/codex-rs/core/tests/suite/agent_execution.rs index f073d7008..69db20cc5 100644 --- a/codex-rs/core/tests/suite/agent_execution.rs +++ b/codex-rs/core/tests/suite/agent_execution.rs @@ -2,7 +2,7 @@ use anyhow::Result; use codex_features::Feature; use core_test_support::responses::ev_assistant_message; use core_test_support::responses::ev_completed; -use core_test_support::responses::ev_function_call; +use core_test_support::responses::ev_function_call_with_namespace; use core_test_support::responses::ev_response_created; use core_test_support::responses::mount_sse_once_match; use core_test_support::responses::sse; @@ -15,6 +15,7 @@ use std::time::Duration; const FIRST_PROMPT: &str = "spawn the first worker"; const FIRST_TASK: &str = "first worker task"; const SECOND_TASK: &str = "second worker task"; +const MULTI_AGENT_V2_NAMESPACE: &str = "collaboration"; fn body_contains(request: &wiremock::Request, text: &str) -> bool { serde_json::from_slice::(&request.body) @@ -47,7 +48,12 @@ async fn v2_nested_spawn_checks_shared_active_execution_capacity() -> Result<()> |request: &wiremock::Request| body_contains(request, FIRST_PROMPT), sse(vec![ ev_response_created("first-response"), - ev_function_call("first-call", "spawn_agent", &first_args), + ev_function_call_with_namespace( + "first-call", + MULTI_AGENT_V2_NAMESPACE, + "spawn_agent", + &first_args, + ), ev_completed("first-response"), ]), ) @@ -63,7 +69,12 @@ async fn v2_nested_spawn_checks_shared_active_execution_capacity() -> Result<()> }, sse(vec![ ev_response_created("first-worker-response"), - ev_function_call("second-call", "spawn_agent", &second_args), + ev_function_call_with_namespace( + "second-call", + MULTI_AGENT_V2_NAMESPACE, + "spawn_agent", + &second_args, + ), ev_completed("first-worker-response"), ]), ) diff --git a/codex-rs/core/tests/suite/model_runtime_selectors.rs b/codex-rs/core/tests/suite/model_runtime_selectors.rs index d893e389d..0c6a87605 100644 --- a/codex-rs/core/tests/suite/model_runtime_selectors.rs +++ b/codex-rs/core/tests/suite/model_runtime_selectors.rs @@ -37,6 +37,7 @@ use tokio::time::sleep; const CHILD_MODEL: &str = "test-multi-agent-child"; const ROOT_MODEL: &str = "test-multi-agent-root"; const ROOT_PROMPT: &str = "spawn a child"; +const MULTI_AGENT_V2_NAMESPACE: &str = "collaboration"; const UNSUPPORTED_CODE_MODE_WARNING: &str = "does not advertise Code Mode support"; struct RemoteModelResponse { @@ -335,7 +336,7 @@ async fn remote_multi_agent_selector_overrides_feature_flags() -> Result<()> { .expect("test config should allow feature update"); }) .await?; - assert!(tool_names(&v2_body).contains(&"send_message".to_string())); + assert!(tool_names(&v2_body).contains(&MULTI_AGENT_V2_NAMESPACE.to_string())); let mut disabled_model = remote_model("test-multi-agent-disabled"); disabled_model.multi_agent_version = Some(MultiAgentVersion::Disabled); @@ -349,7 +350,12 @@ async fn remote_multi_agent_selector_overrides_feature_flags() -> Result<()> { let disabled_tools = tool_names(&disabled_body); assert!(disabled_tools.iter().all(|name| !matches!( name.as_str(), - "multi_agent_v1" | "spawn_agent" | "send_message" | "wait_agent" | "list_agents" + "multi_agent_v1" + | MULTI_AGENT_V2_NAMESPACE + | "spawn_agent" + | "send_message" + | "wait_agent" + | "list_agents" ))); Ok(()) @@ -432,7 +438,7 @@ async fn remote_multi_agent_selector_uses_model_selected_before_first_turn() -> .expect("expected response request") .body_json(), ) - .contains(&"send_message".to_string()), + .contains(&MULTI_AGENT_V2_NAMESPACE.to_string()), ), (1, Some(MultiAgentVersion::V2), true) ); diff --git a/codex-rs/core/tests/suite/pending_input.rs b/codex-rs/core/tests/suite/pending_input.rs index 269beae63..17e00513f 100644 --- a/codex-rs/core/tests/suite/pending_input.rs +++ b/codex-rs/core/tests/suite/pending_input.rs @@ -20,6 +20,7 @@ use core_test_support::responses; use core_test_support::responses::ev_completed; use core_test_support::responses::ev_completed_with_tokens; use core_test_support::responses::ev_function_call; +use core_test_support::responses::ev_function_call_with_namespace; use core_test_support::responses::ev_message_item_added; use core_test_support::responses::ev_output_text_delta; use core_test_support::responses::ev_reasoning_item; @@ -290,11 +291,13 @@ async fn steer_interrupts_wait_agent_and_is_sent_in_follow_up_request() { const WAIT_CALL_ID: &str = "wait-call"; const INITIAL_PROMPT: &str = "wait for an agent"; const STEER_PROMPT: &str = "stop waiting and continue"; + const MULTI_AGENT_V2_NAMESPACE: &str = "collaboration"; let first_chunks = vec![ chunk(ev_response_created("resp-1")), - chunk(ev_function_call( + chunk(ev_function_call_with_namespace( WAIT_CALL_ID, + MULTI_AGENT_V2_NAMESPACE, "wait_agent", r#"{"timeout_ms":10000}"#, )), diff --git a/codex-rs/core/tests/suite/rollout_budget.rs b/codex-rs/core/tests/suite/rollout_budget.rs index c051f01af..6aa211bcc 100644 --- a/codex-rs/core/tests/suite/rollout_budget.rs +++ b/codex-rs/core/tests/suite/rollout_budget.rs @@ -10,7 +10,7 @@ use core_test_support::responses::ResponsesRequest; use core_test_support::responses::ev_assistant_message; use core_test_support::responses::ev_completed; use core_test_support::responses::ev_completed_with_tokens; -use core_test_support::responses::ev_function_call; +use core_test_support::responses::ev_function_call_with_namespace; use core_test_support::responses::ev_response_created; use core_test_support::responses::mount_sse_once_match; use core_test_support::responses::mount_sse_sequence; @@ -25,6 +25,8 @@ use std::time::Duration; use test_case::test_case; use tokio::time::timeout; +const MULTI_AGENT_V2_NAMESPACE: &str = "collaboration"; + fn rollout_budget() -> RolloutBudgetConfig { RolloutBudgetConfig { limit_tokens: 100, @@ -130,7 +132,12 @@ async fn subagent_usage_draws_from_the_shared_budget() -> Result<()> { |request: &wiremock::Request| wire_request_contains(request, ROOT_PROMPT), sse(vec![ ev_response_created("root-1"), - ev_function_call(SPAWN_CALL_ID, "spawn_agent", &spawn_args), + ev_function_call_with_namespace( + SPAWN_CALL_ID, + MULTI_AGENT_V2_NAMESPACE, + "spawn_agent", + &spawn_args, + ), ev_completed_with_tokens("root-1", /*total_tokens*/ 10), ]), ) diff --git a/codex-rs/core/tests/suite/subagent_notifications.rs b/codex-rs/core/tests/suite/subagent_notifications.rs index c49e47496..53f9a97bc 100644 --- a/codex-rs/core/tests/suite/subagent_notifications.rs +++ b/codex-rs/core/tests/suite/subagent_notifications.rs @@ -17,7 +17,6 @@ use core_test_support::hooks::trust_discovered_hooks; use core_test_support::responses::ResponsesRequest; use core_test_support::responses::ev_assistant_message; use core_test_support::responses::ev_completed; -use core_test_support::responses::ev_function_call; use core_test_support::responses::ev_function_call_with_namespace; use core_test_support::responses::ev_response_created; use core_test_support::responses::ev_tool_search_call; @@ -48,6 +47,7 @@ use wiremock::MockServer; const SPAWN_CALL_ID: &str = "spawn-call-1"; const MULTI_AGENT_V1_NAMESPACE: &str = "multi_agent_v1"; +const MULTI_AGENT_V2_NAMESPACE: &str = "collaboration"; const TURN_0_FORK_PROMPT: &str = "seed fork context"; const TURN_1_PROMPT: &str = "spawn a child and continue"; const TURN_2_NO_WAIT_PROMPT: &str = "follow up without wait"; @@ -63,6 +63,23 @@ const SUBAGENT_STOP_CONTINUATION: &str = "continue only the child"; const INTERNAL_SUBAGENT_PROMPT: &str = "internal subagent: review"; fn body_contains(req: &wiremock::Request, text: &str) -> bool { + decoded_body(req) + .and_then(|body| String::from_utf8(body).ok()) + .is_some_and(|body| body.contains(text)) +} + +fn request_has_input_type(req: &wiremock::Request, ty: &str) -> bool { + decoded_body(req) + .and_then(|body| serde_json::from_slice::(&body).ok()) + .and_then(|body| body.get("input").and_then(Value::as_array).cloned()) + .is_some_and(|items| { + items + .iter() + .any(|item| item.get("type").and_then(Value::as_str) == Some(ty)) + }) +} + +fn decoded_body(req: &wiremock::Request) -> Option> { let is_zstd = req .headers .get("content-encoding") @@ -72,14 +89,11 @@ fn body_contains(req: &wiremock::Request, text: &str) -> bool { .split(',') .any(|entry| entry.trim().eq_ignore_ascii_case("zstd")) }); - let bytes = if is_zstd { + if is_zstd { zstd::stream::decode_all(std::io::Cursor::new(&req.body)).ok() } else { Some(req.body.clone()) - }; - bytes - .and_then(|body| String::from_utf8(body).ok()) - .is_some_and(|body| body.contains(text)) + } } fn has_subagent_notification(req: &ResponsesRequest) -> bool { @@ -1040,14 +1054,19 @@ async fn encrypted_multi_agent_v2_spawn_sends_agent_message_to_child() -> Result |req: &wiremock::Request| body_contains(req, TURN_1_PROMPT), sse(vec![ ev_response_created("resp-parent-1"), - ev_function_call(SPAWN_CALL_ID, "spawn_agent", &spawn_args), + ev_function_call_with_namespace( + SPAWN_CALL_ID, + MULTI_AGENT_V2_NAMESPACE, + "spawn_agent", + &spawn_args, + ), ev_completed("resp-parent-1"), ]), ) .await; let child_request_log = mount_sse_once_match( &server, - |req: &wiremock::Request| body_contains(req, "\"type\":\"agent_message\""), + |req: &wiremock::Request| request_has_input_type(req, "agent_message"), sse(vec![ ev_response_created("resp-child-1"), ev_completed("resp-child-1"), @@ -1057,7 +1076,7 @@ async fn encrypted_multi_agent_v2_spawn_sends_agent_message_to_child() -> Result mount_sse_once_match( &server, |req: &wiremock::Request| { - body_contains(req, SPAWN_CALL_ID) && !body_contains(req, "\"type\":\"agent_message\"") + body_contains(req, SPAWN_CALL_ID) && !request_has_input_type(req, "agent_message") }, sse(vec![ ev_response_created("resp-parent-2"), @@ -1129,7 +1148,12 @@ async fn plaintext_multi_agent_v2_completion_sends_agent_message( |req: &wiremock::Request| body_contains(req, TURN_1_PROMPT), sse(vec![ ev_response_created("resp-parent-1"), - ev_function_call(SPAWN_CALL_ID, "spawn_agent", &spawn_args), + ev_function_call_with_namespace( + SPAWN_CALL_ID, + MULTI_AGENT_V2_NAMESPACE, + "spawn_agent", + &spawn_args, + ), ev_completed("resp-parent-1"), ]), ) @@ -1144,7 +1168,7 @@ async fn plaintext_multi_agent_v2_completion_sends_agent_message( }; let child_request = mount_response_once_match( &server, - |req: &wiremock::Request| body_contains(req, "\"type\":\"agent_message\""), + |req: &wiremock::Request| request_has_input_type(req, "agent_message"), sse_response(sse(child_events)).set_delay(Duration::from_secs(1)), ) .await; @@ -1183,7 +1207,12 @@ async fn plaintext_multi_agent_v2_completion_sends_agent_message( }, sse(vec![ ev_response_created("resp-parent-3"), - ev_function_call("wait-agent-call", "wait_agent", "{}"), + ev_function_call_with_namespace( + "wait-agent-call", + MULTI_AGENT_V2_NAMESPACE, + "wait_agent", + "{}", + ), ev_completed("resp-parent-3"), ]), )