dynamic tool calls: add param exposeToContext to optionally hide tool (#14501)

This extends dynamic_tool_calls to allow us to hide a tool from the
model context but still use it as part of the general tool calling
runtime (for ex from js_repl/code_mode)
This commit is contained in:
Channing Conger
2026-03-14 01:58:43 -07:00
committed by GitHub
parent e389091042
commit 70eddad6b0
16 changed files with 578 additions and 14 deletions
+17 -1
View File
@@ -6219,9 +6219,25 @@ fn build_prompt(
turn_context: &TurnContext,
base_instructions: BaseInstructions,
) -> Prompt {
let deferred_dynamic_tools = turn_context
.dynamic_tools
.iter()
.filter(|tool| tool.defer_loading)
.map(|tool| tool.name.as_str())
.collect::<HashSet<_>>();
let tools = if deferred_dynamic_tools.is_empty() {
router.model_visible_specs()
} else {
router
.model_visible_specs()
.into_iter()
.filter(|spec| !deferred_dynamic_tools.contains(spec.name()))
.collect()
};
Prompt {
input,
tools: router.model_visible_specs(),
tools,
parallel_tool_calls: turn_context.model_info.supports_parallel_tool_calls,
base_instructions,
personality: turn_context.personality,
@@ -1851,6 +1851,7 @@ async fn js_repl_emit_image_rejects_mixed_content() -> anyhow::Result<()> {
"properties": {},
"additionalProperties": false
}),
defer_loading: false,
}])
.await;
if !turn
@@ -1949,6 +1950,7 @@ async fn js_repl_dynamic_tool_response_preserves_js_line_separator_text() -> any
"properties": {},
"additionalProperties": false
}),
defer_loading: false,
}])
.await;
@@ -2008,6 +2010,79 @@ console.log(text);
Ok(())
}
#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
async fn js_repl_can_call_hidden_dynamic_tools() -> anyhow::Result<()> {
if !can_run_js_repl_runtime_tests().await {
return Ok(());
}
let (session, turn, rx_event) =
make_session_and_context_with_dynamic_tools_and_rx(vec![DynamicToolSpec {
name: "hidden_dynamic_tool".to_string(),
description: "A hidden dynamic tool.".to_string(),
input_schema: serde_json::json!({
"type": "object",
"properties": {
"city": { "type": "string" }
},
"required": ["city"],
"additionalProperties": false
}),
defer_loading: true,
}])
.await;
*session.active_turn.lock().await = Some(crate::state::ActiveTurn::default());
let tracker = Arc::new(tokio::sync::Mutex::new(TurnDiffTracker::default()));
let manager = turn.js_repl.manager().await?;
let code = r#"
const out = await codex.tool("hidden_dynamic_tool", { city: "Paris" });
console.log(JSON.stringify(out));
"#;
let session_for_response = Arc::clone(&session);
let response_watcher = async move {
loop {
let event = tokio::time::timeout(Duration::from_secs(2), rx_event.recv()).await??;
if let EventMsg::DynamicToolCallRequest(request) = event.msg {
session_for_response
.notify_dynamic_tool_response(
&request.call_id,
DynamicToolResponse {
content_items: vec![DynamicToolCallOutputContentItem::InputText {
text: "hidden-ok".to_string(),
}],
success: true,
},
)
.await;
return Ok::<(), anyhow::Error>(());
}
}
};
let (result, response_watcher_result) = tokio::join!(
manager.execute(
Arc::clone(&session),
Arc::clone(&turn),
tracker,
JsReplArgs {
code: code.to_string(),
timeout_ms: Some(15_000),
},
),
response_watcher,
);
let result = result?;
response_watcher_result?;
assert!(result.output.contains("hidden-ok"));
assert!(session.get_pending_input().await.is_empty());
Ok(())
}
#[tokio::test]
async fn js_repl_prefers_env_node_module_dirs_over_config() -> anyhow::Result<()> {
if !can_run_js_repl_runtime_tests().await {
+204 -7
View File
@@ -4,6 +4,14 @@ use anyhow::Result;
use codex_core::config::types::McpServerConfig;
use codex_core::config::types::McpServerTransportConfig;
use codex_core::features::Feature;
use codex_protocol::dynamic_tools::DynamicToolCallOutputContentItem;
use codex_protocol::dynamic_tools::DynamicToolResponse;
use codex_protocol::dynamic_tools::DynamicToolSpec;
use codex_protocol::protocol::AskForApproval;
use codex_protocol::protocol::EventMsg;
use codex_protocol::protocol::Op;
use codex_protocol::protocol::SandboxPolicy;
use codex_protocol::user_input::UserInput;
use core_test_support::assert_regex_match;
use core_test_support::responses;
use core_test_support::responses::ResponseMock;
@@ -17,6 +25,8 @@ use core_test_support::skip_if_no_network;
use core_test_support::stdio_server_bin;
use core_test_support::test_codex::TestCodex;
use core_test_support::test_codex::test_codex;
use core_test_support::wait_for_event;
use core_test_support::wait_for_event_match;
use pretty_assertions::assert_eq;
use serde_json::Value;
use std::collections::HashMap;
@@ -91,18 +101,39 @@ fn custom_tool_output_body_and_success(
req: &ResponsesRequest,
call_id: &str,
) -> (String, Option<bool>) {
let (_, success) = req
let (content, success) = req
.custom_tool_call_output_content_and_success(call_id)
.expect("custom tool output should be present");
let items = custom_tool_output_items(req, call_id);
let output = items
let text_items = items
.iter()
.skip(1)
.filter_map(|item| item.get("text").and_then(Value::as_str))
.collect();
.collect::<Vec<_>>();
let output = match text_items.as_slice() {
[] => content.unwrap_or_default(),
[only] => (*only).to_string(),
[_, rest @ ..] => rest.concat(),
};
(output, success)
}
fn custom_tool_output_last_non_empty_text(req: &ResponsesRequest, call_id: &str) -> Option<String> {
match req.custom_tool_call_output(call_id).get("output") {
Some(Value::String(text)) if !text.trim().is_empty() => Some(text.clone()),
Some(Value::Array(items)) => items
.iter()
.filter_map(|item| item.get("text").and_then(Value::as_str))
.rfind(|text| !text.trim().is_empty())
.map(str::to_string),
Some(Value::String(_))
| Some(Value::Object(_))
| Some(Value::Number(_))
| Some(Value::Bool(_))
| Some(Value::Null)
| None => None,
}
}
async fn run_code_mode_turn(
server: &MockServer,
prompt: &str,
@@ -1506,6 +1537,10 @@ text({ json: true });
let req = second_mock.single_request();
let (output, success) = custom_tool_output_body_and_success(&req, "call-1");
eprintln!(
"hidden dynamic tool raw output: {}",
req.custom_tool_call_output("call-1")
);
assert_ne!(
success,
Some(false),
@@ -1920,7 +1955,10 @@ text(JSON.stringify(tool));
"exec ALL_TOOLS lookup failed unexpectedly: {output}"
);
let parsed: Value = serde_json::from_str(&output)?;
let parsed: Value = serde_json::from_str(
&custom_tool_output_last_non_empty_text(&req, "call-1")
.expect("exec ALL_TOOLS lookup should emit JSON"),
)?;
assert_eq!(
parsed,
serde_json::json!({
@@ -1955,7 +1993,10 @@ text(JSON.stringify(tool));
"exec ALL_TOOLS MCP lookup failed unexpectedly: {output}"
);
let parsed: Value = serde_json::from_str(&output)?;
let parsed: Value = serde_json::from_str(
&custom_tool_output_last_non_empty_text(&req, "call-1")
.expect("exec ALL_TOOLS MCP lookup should emit JSON"),
)?;
assert_eq!(
parsed,
serde_json::json!({
@@ -1967,6 +2008,159 @@ text(JSON.stringify(tool));
Ok(())
}
#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
async fn code_mode_can_call_hidden_dynamic_tools() -> Result<()> {
skip_if_no_network!(Ok(()));
let server = responses::start_mock_server().await;
let mut builder = test_codex().with_config(move |config| {
let _ = config.features.enable(Feature::CodeMode);
});
let base_test = builder.build(&server).await?;
let new_thread = base_test
.thread_manager
.start_thread_with_tools(
base_test.config.clone(),
vec![DynamicToolSpec {
name: "hidden_dynamic_tool".to_string(),
description: "A hidden dynamic tool.".to_string(),
input_schema: serde_json::json!({
"type": "object",
"properties": {
"city": { "type": "string" }
},
"required": ["city"],
"additionalProperties": false,
}),
defer_loading: true,
}],
false,
)
.await?;
let test = TestCodex {
home: base_test.home,
cwd: base_test.cwd,
codex: new_thread.thread,
session_configured: new_thread.session_configured,
config: base_test.config,
thread_manager: base_test.thread_manager,
};
let code = r#"
import { ALL_TOOLS, hidden_dynamic_tool } from "tools.js";
const tool = ALL_TOOLS.find(({ name }) => name === "hidden_dynamic_tool");
const out = await hidden_dynamic_tool({ city: "Paris" });
text(
JSON.stringify({
name: tool?.name ?? null,
description: tool?.description ?? null,
out,
})
);
"#;
responses::mount_sse_once(
&server,
sse(vec![
ev_response_created("resp-1"),
ev_custom_tool_call("call-1", "exec", code),
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.codex
.submit(Op::UserTurn {
items: vec![UserInput::Text {
text: "use exec to inspect and call hidden tools".into(),
text_elements: Vec::new(),
}],
final_output_json_schema: None,
cwd: test.cwd.path().to_path_buf(),
approval_policy: AskForApproval::Never,
sandbox_policy: SandboxPolicy::DangerFullAccess,
model: test.session_configured.model.clone(),
effort: None,
summary: None,
service_tier: None,
collaboration_mode: None,
personality: None,
})
.await?;
let turn_id = wait_for_event_match(&test.codex, |event| match event {
EventMsg::TurnStarted(event) => Some(event.turn_id.clone()),
_ => None,
})
.await;
let request = wait_for_event_match(&test.codex, |event| match event {
EventMsg::DynamicToolCallRequest(request) => Some(request.clone()),
_ => None,
})
.await;
assert_eq!(request.tool, "hidden_dynamic_tool");
assert_eq!(request.arguments, serde_json::json!({ "city": "Paris" }));
test.codex
.submit(Op::DynamicToolResponse {
id: request.call_id,
response: DynamicToolResponse {
content_items: vec![DynamicToolCallOutputContentItem::InputText {
text: "hidden-ok".to_string(),
}],
success: true,
},
})
.await?;
wait_for_event(&test.codex, |event| match event {
EventMsg::TurnComplete(event) => event.turn_id == turn_id,
_ => false,
})
.await;
let req = second_mock.single_request();
let (output, success) = custom_tool_output_body_and_success(&req, "call-1");
assert_ne!(
success,
Some(false),
"exec hidden dynamic tool call failed unexpectedly: {output}"
);
let parsed: Value = serde_json::from_str(
&custom_tool_output_last_non_empty_text(&req, "call-1")
.expect("exec hidden dynamic tool lookup should emit JSON"),
)?;
assert_eq!(
parsed.get("name"),
Some(&Value::String("hidden_dynamic_tool".to_string()))
);
assert_eq!(
parsed.get("out"),
Some(&Value::String("hidden-ok".to_string()))
);
assert!(
parsed
.get("description")
.and_then(Value::as_str)
.is_some_and(|description| {
description.contains("A hidden dynamic tool.")
&& description.contains("declare const tools:")
&& description.contains("hidden_dynamic_tool(args:")
})
);
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(()));
@@ -2130,7 +2324,10 @@ text(JSON.stringify(load("nb")));
Some(false),
"exec load call failed unexpectedly: {second_output}"
);
let loaded: Value = serde_json::from_str(&second_output)?;
let loaded: Value = serde_json::from_str(
&custom_tool_output_last_non_empty_text(&second_request, "call-2")
.expect("exec load call should emit JSON"),
)?;
assert_eq!(
loaded,
serde_json::json!({ "title": "Notebook", "items": [1, true, null] })
@@ -110,6 +110,7 @@ async fn backfill_scans_existing_rollouts() -> Result<()> {
"required": ["city"],
"properties": { "city": { "type": "string" } }
}),
defer_loading: true,
},
DynamicToolSpec {
name: "weather_lookup".to_string(),
@@ -119,6 +120,7 @@ async fn backfill_scans_existing_rollouts() -> Result<()> {
"required": ["zip"],
"properties": { "zip": { "type": "string" } }
}),
defer_loading: false,
},
];
let dynamic_tools_for_hook = dynamic_tools.clone();