[codex] Update realtime V2 VAD silence delay and 1.5 prompt (#18092)

## Summary

- set the realtime v2 server VAD silence delay to 500ms
- update the default realtime 1.5 backend prompt to the v4 text
- keep the session payload and prompt rendering tests aligned with those
changes

## Why

- the VAD change gives the voice path a longer pause before ending the
user's turn
- the prompt change makes the default bundled realtime prompt match the
current v4 content

## Validation

- `cargo +1.93.0 test -p codex-core realtime_prompt --manifest-path
/tmp/codex-realtime-v2-vad-prompt-v4/codex-rs/Cargo.toml`
- `CARGO_TARGET_DIR=/tmp/codex-pr-v4-target cargo +1.93.0 test -p
codex-api
realtime_v2_session_update_includes_background_agent_tool_and_handoff_output_item
--manifest-path
/tmp/codex-realtime-v2-vad-prompt-v4/codex-rs/Cargo.toml`
- `CARGO_TARGET_DIR=/tmp/codex-pr-v4-target cargo +1.93.0 test -p
codex-app-server --test all
'suite::v2::realtime_conversation::realtime_webrtc_start_emits_sdp_notification'
--manifest-path /tmp/codex-realtime-v2-vad-prompt-v4/codex-rs/Cargo.toml
-- --exact`
This commit is contained in:
bxie-openai
2026-04-16 14:30:57 -07:00
committed by GitHub
parent d9c71d41a9
commit 6a1ddfc366
6 changed files with 49 additions and 44 deletions

View File

@@ -1142,7 +1142,7 @@ async fn realtime_webrtc_start_emits_sdp_notification() -> Result<()> {
Some("multipart/form-data; boundary=codex-realtime-call-boundary")
);
let body = String::from_utf8(request.body).context("multipart body should be utf-8")?;
let session = r#"{"tool_choice":"auto","type":"realtime","model":"gpt-realtime-1.5","instructions":"backend prompt\n\nstartup context","output_modalities":["audio"],"audio":{"input":{"format":{"type":"audio/pcm","rate":24000},"noise_reduction":{"type":"near_field"},"turn_detection":{"type":"server_vad","interrupt_response":true,"create_response":true}},"output":{"format":{"type":"audio/pcm","rate":24000},"voice":"marin"}},"tools":[{"type":"function","name":"background_agent","description":"Send a user request to the background agent. Use this as the default action. Do not rephrase the user's ask or rewrite it in your own words; pass along the user's own words. If the background agent is idle, this starts a new task and returns the final result to the user. If the background agent is already working on a task, this sends the request as guidance to steer that previous task. If the user asks to do something next, later, after this, or once current work finishes, call this tool so the work is actually queued instead of merely promising to do it later.","parameters":{"type":"object","properties":{"prompt":{"type":"string","description":"The user request to delegate to the background agent."}},"required":["prompt"],"additionalProperties":false}}]}"#;
let session = r#"{"tool_choice":"auto","type":"realtime","model":"gpt-realtime-1.5","instructions":"backend prompt\n\nstartup context","output_modalities":["audio"],"audio":{"input":{"format":{"type":"audio/pcm","rate":24000},"noise_reduction":{"type":"near_field"},"turn_detection":{"type":"server_vad","interrupt_response":true,"create_response":true,"silence_duration_ms":500}},"output":{"format":{"type":"audio/pcm","rate":24000},"voice":"marin"}},"tools":[{"type":"function","name":"background_agent","description":"Send a user request to the background agent. Use this as the default action. Do not rephrase the user's ask or rewrite it in your own words; pass along the user's own words. If the background agent is idle, this starts a new task and returns the final result to the user. If the background agent is already working on a task, this sends the request as guidance to steer that previous task. If the user asks to do something next, later, after this, or once current work finishes, call this tool so the work is actually queued instead of merely promising to do it later.","parameters":{"type":"object","properties":{"prompt":{"type":"string","description":"The user request to delegate to the background agent."}},"required":["prompt"],"additionalProperties":false}}]}"#;
let session = normalized_json_string(session)?;
assert_eq!(
body,