openai/codex
1
0
Fork 0
mirror of https://github.com/openai/codex.git synced 2026-09-06 15:29:32 +00:00
Code Issues Packages Projects Releases Wiki Activity

Restrict async user messages to questions (#41365)

Browse Source 
## What changed

Update the `send_user_message_async` tool guidance to reserve it for questions
that require missing information, preferences, constraints, clarification, or
approval. Ask models to keep these questions concise, understandable, and at
an appropriate level of abstraction for the user and task.

GitOrigin-RevId: 4a16754401fb5fc106bbf65759ac4a7a78b6c7b3
This commit is contained in:
rhan-oai
2026-08-28 16:33:48 +00:00
committed by copyberry
parent 8faf7252f0
commit 7e41be641e
1 changed files with 2 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
codex-rs/core/src/tools/handlers/send_user_message_async.rs
View File

@@ -37,13 +37,13 @@ impl ToolExecutor<ToolInvocation> for SendUserMessageAsyncHandler {
let properties = BTreeMap::from([(
"message".to_string(),
JsonSchema::string(Some(
"The concise question or update to send to the user.".to_string(),
"The concise question to send to the user.".to_string(),
)),
)]);
ToolSpec::Function(ResponsesApiTool {
name: TOOL_NAME.to_string(),
description: "Send a concise message that needs the user's attention during ongoing work. The tool returns immediately without ending the turn or waiting for a reply; any reply arrives asynchronously as a new user message. Use this tool to ask for missing information, preferences, constraints, clarification, or approval; report a critical blocker or a finding that may change the task's direction; or answer a user question or status request received while work is still in progress. Use this tool when a message needs the user's immediate attention; use commentary for routine progress and intermediate context. Use clear formatting, such as bolding questions, to make requests easy to notice and answer."
description: "Send a concise message that needs the user's attention during ongoing work. The tool returns immediately without ending the turn or waiting for a reply; any reply arrives asynchronously as a new user message.\nOnly use this tool to ask for missing information, preferences, constraints, clarification, or approval. The message should be concise, easy to read and understand, and at the right level of abstraction that is appropriate for the user and task at hand."
.to_string(),
strict: false,
defer_loading: None,