code-mode: add process-owned session runtime by cconger · Pull Request #29805 · openai/codex

cconger · 2026-06-24T05:25:15Z

Why

Code mode needs a process-owned implementation that preserves the existing CodeModeSession contract while moving JavaScript execution and stored session values out of the client process. The host and client are introduced together here so the transport is reviewable as one end-to-end path rather than as disconnected process halves.

What changed

adds the codex-code-mode-host sidecar binary
lazily spawns and negotiates one process host for a provider
multiplexes logical sessions and request/response correlation over framed stdin/stdout
implements session open, execute, wait, terminate, and shutdown
hosts each logical session with InProcessCodeModeSession
forwards nested tool calls and notifications through a remote delegate
transports the delayed initial runtime response separately from execution admission
shuts hosted sessions down when the connection ends
adds end-to-end coverage for stored values, delegates, and cell control

This does not activate the process-owned provider in the production tool path. Packaging and binary discovery remain follow-up work.

Testing

just test -p codex-code-mode
just test -p codex-code-mode-host

Stack

Part 2 of 6. Depends on #29804; followed by #29806.

jif-oai · 2026-06-24T11:59:00Z

+            self.state.pending.lock().await.remove(&id);
+            return Err(err);
+        }
+        response_rx


This isn’t cancellation-safe: once the frame is sent, dropping the caller only drops response_rx

jif-oai · 2026-06-24T11:59:30Z

+                        let Some(message) = message else {
+                            break;
+                        };
+                        if let Err(err) = writer.write(&message).await {


This makes an oversized client→host message connection-fatal. A nested-tool result can exceed the frame cap, so one large MCP/image/exec result tears down every session sharing this host
Is this on purpose? (not sure what's the latest decision here)

jif-oai · 2026-06-24T11:59:43Z

+        for (_, sender) in self.initial_responses.lock().await.drain() {
+            let _ = sender.send(Err(reason.clone()));
+        }
+        self.delegates.lock().await.clear();


Clearing the delegates skips cell_closed for every live remote cell

jif-oai · 2026-06-24T12:00:12Z

+        let response = self
+            .request(HostRequest::OpenSession {
+                session_id: session_id.clone(),
+            })


There’s an open-session leak here if create_session() is cancelled: neither cleanup arm runs, so the host can retain the session and this map retains the delegate after the only local handle is gone

jif-oai · 2026-06-24T12:00:30Z

+                .state
+                .lock()
+                .unwrap_or_else(std::sync::PoisonError::into_inner);
+            match std::mem::replace(&mut *state, SessionState::Shutdown) {


Shutdown is published before the host acknowledges cleanup... this is race prone

jif-oai · 2026-06-24T12:01:01Z

+            .write(&ClientToHost::ClientHello(hello))
+            .await
+            .map_err(|err| format!("failed to write code-mode host hello: {err}"))?;
+        match reader


Can we put the startup hello exchange behind a finite deadline and start draining stderr before awaiting it?

jif-oai · 2026-06-24T12:11:09Z

+    });
+    let writer_task = tokio::spawn(async move {
+        while let Some(message) = outgoing_rx.recv().await {
+            writer


(found by Codex)
The reverse direction has the same blast radius: JS can build an arbitrarily large tool argument or notify() string, which becomes one DelegateRequest. If it exceeds MAX_FRAME_BYTES, this stops the writer and tears down the shared host

## Why The process-owned code mode implementation needs an explicit, bounded wire contract before either side depends on it. Keeping framing and message semantics in `codex-code-mode-protocol` gives the client and sidecar one shared source of truth and makes compatibility failures detectable during connection setup. ## What changed - adds a versioned client/host handshake with required and optional capabilities - defines operation requests and responses for session lifecycle and cell control - defines reverse delegate request, response, cancellation, and cell-closure messages - adds a four-byte little-endian length-prefixed JSON codec with a hard frame cap - rejects malformed frames, unknown fields, invalid identifiers, and unsupported protocol states - locks the wire representation down with explicit JSON round-trip tests ## Testing - `just test -p codex-code-mode-protocol` ## Stack Part 1 of 6. Followed by [#29805](#29805).

## Why The process-owned code mode implementation needs an explicit, bounded wire contract before either side depends on it. Keeping framing and message semantics in `codex-code-mode-protocol` gives the client and sidecar one shared source of truth and makes compatibility failures detectable during connection setup. ## What changed - adds a versioned client/host handshake with required and optional capabilities - defines operation requests and responses for session lifecycle and cell control - defines reverse delegate request, response, cancellation, and cell-closure messages - adds a four-byte little-endian length-prefixed JSON codec with a hard frame cap - rejects malformed frames, unknown fields, invalid identifiers, and unsupported protocol states - locks the wire representation down with explicit JSON round-trip tests ## Testing - `just test -p codex-code-mode-protocol` ## Stack Part 1 of 6. Followed by [openai#29805](openai#29805).

This was referenced Jun 24, 2026

code-mode: define process host wire protocol #29804

Merged

code-mode: cover remote session runtime behavior #29806

Closed

cconger added the code-reviewed label Jun 24, 2026

jif-oai reviewed Jun 24, 2026

View reviewed changes

cconger force-pushed the cconger/process-owned-session-runtime branch 2 times, most recently from 0f7bda8 to 537cda9 Compare June 24, 2026 21:45

cconger force-pushed the cconger/process-owned-session-protocol branch from e0fac5a to 125a59b Compare June 24, 2026 22:26

cconger force-pushed the cconger/process-owned-session-runtime branch from 537cda9 to 42691b9 Compare June 24, 2026 22:26

cconger added 6 commits June 25, 2026 01:00

Define code mode host wire protocol

020ecd3

Pin code mode host V1 payloads

884e77a

Strengthen code mode wire integer types

077f819

code-mode: satisfy argument comment lint

c52c04b

Add process-owned code mode session

464273e

code-mode: annotate host request ids

1886ad6

cconger force-pushed the cconger/process-owned-session-protocol branch from 125a59b to c52c04b Compare June 25, 2026 01:08

cconger force-pushed the cconger/process-owned-session-runtime branch from 42691b9 to 1886ad6 Compare June 25, 2026 01:08

Base automatically changed from cconger/process-owned-session-protocol to main June 25, 2026 03:03

github-actions Bot mentioned this pull request Jun 25, 2026

📊 AI CLI 工具社区动态日报 2026-06-25 96loveslife/big_model_radar#54

Open

cconger closed this Jun 26, 2026

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

code-mode: add process-owned session runtime - #29805

code-mode: add process-owned session runtime#29805
cconger wants to merge 6 commits into
mainfrom
cconger/process-owned-session-runtime

cconger commented Jun 24, 2026 •

edited

Loading

Uh oh!

jif-oai Jun 24, 2026

Uh oh!

jif-oai Jun 24, 2026

Uh oh!

jif-oai Jun 24, 2026

Uh oh!

jif-oai Jun 24, 2026

Uh oh!

jif-oai Jun 24, 2026

Uh oh!

jif-oai Jun 24, 2026

Uh oh!

jif-oai Jun 24, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

cconger commented Jun 24, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Why

What changed

Testing

Stack

Uh oh!

jif-oai Jun 24, 2026

Choose a reason for hiding this comment

Uh oh!

jif-oai Jun 24, 2026

Choose a reason for hiding this comment

Uh oh!

jif-oai Jun 24, 2026

Choose a reason for hiding this comment

Uh oh!

jif-oai Jun 24, 2026

Choose a reason for hiding this comment

Uh oh!

jif-oai Jun 24, 2026

Choose a reason for hiding this comment

Uh oh!

jif-oai Jun 24, 2026

Choose a reason for hiding this comment

Uh oh!

jif-oai Jun 24, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

cconger commented Jun 24, 2026 •

edited

Loading