Add sticky environment API and thread state by starr-openai · Pull Request #18897 · openai/codex

starr-openai · 2026-04-21T21:55:31Z

Summary

add sticky environment selections to app-server v2 thread/start and turn/start request flow
carry thread-level selections through core session/thread state
add app-server coverage for sticky selections and turn overrides

Stack

This PR: API and thread persistence
Load named environments from config #18898: config.toml named environment loading
Wire selected environments into tools #18899: downstream tool/runtime consumers

Validation

Not run locally; split only.

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 606663dfae

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

Allow thread/start to configure sticky environment selections that are used by turns when no per-turn override is supplied. Per-turn environments continue to take precedence, while omitted thread selections preserve the existing default behavior. Co-authored-by: Codex <noreply@openai.com>

Cover sticky thread environment selections and turn-level overrides through the app-server v2 thread/start and turn/start JSON-RPC flow. The matrix mirrors the manual smoke cases for omitted, empty, local, remote, and local plus remote selections. Co-authored-by: Codex <noreply@openai.com>

Group thread-start options for lint-friendly callsites and update generated v2 schema for sticky environment selections. Co-authored-by: Codex <noreply@openai.com>

Introduce a core-owned EnvironmentSelection type so app-server converts API environment params at the core boundary instead of passing protocol operation structs through session/thread state. Rename the internal sticky field from environment_selections to environments to match the v2 API shape. Co-authored-by: Codex <noreply@openai.com>

Reject named turn and sticky environments when environment access is disabled, while preserving explicit empty environment selections as no-environment turns. Validate sticky environment state before storing it so invalid selections cannot poison future turns. Co-authored-by: Codex <noreply@openai.com>

Co-authored-by: Codex <noreply@openai.com>

Resume and fork should not recover thread environment overrides from persisted rollout state; sticky environment selection stays scoped to the live thread/session configuration. Co-authored-by: Codex <noreply@openai.com>

Remove the duplicate SessionConfiguration environments initializer left by the rebase. Co-authored-by: Codex <noreply@openai.com>

Avoid requiring TurnEnvironment to implement PartialEq in the resume/fork test that verifies sticky environments are not restored from rollout history. Co-authored-by: Codex <noreply@openai.com>

Co-authored-by: Codex <noreply@openai.com>

Move default thread environment selection into a shared helper, validate explicit environment IDs at spawn, and resolve selections only when constructing turns. This keeps app-server optionality at the API edge while preserving sticky thread state as explicit selections. Co-authored-by: Codex <noreply@openai.com>

Use the EnvironmentManager default environment id when materializing default thread selections, carry environment ids through resolved turn environments, and make subagents inherit the parent turn environment list instead of re-defaulting. Keep turn-local environment overrides on SessionSettingsUpdate so turn context updates flow through one settings payload. Co-authored-by: Codex <noreply@openai.com>

Keep the effective turn environment selections on TurnContext so child agents can inherit selections directly without rebuilding them from resolved environments. Route app-server validation through ThreadManager and preserve turn-scoped selections for agent job workers. Co-authored-by: Codex <noreply@openai.com>

Remove the unused resolved turn environment id field now that effective selections are stored separately, and collapse app-server environment validation guards for clippy. Co-authored-by: Codex <noreply@openai.com>

pakrym-oai · 2026-04-23T22:03:34Z

 pub(crate) struct SpawnAgentOptions {
    pub(crate) fork_parent_spawn_call_id: Option<String>,
    pub(crate) fork_mode: Option<SpawnAgentForkMode>,
+    pub(crate) environments: Option<Vec<TurnEnvironmentSelection>>,


nit: non optional? non blocking

this one still need Option for the 'use defaults' mode, which a few other callsites still use (e.g. memories/phase2.rs)

pakrym-oai · 2026-04-23T22:05:17Z

        reasoning_summary,
        session_source,
        environment: parent_turn_context.environment.clone(),
+        environment_selections: parent_turn_context.environment_selections.clone(),


nit: why are we storing both environment_selections and environments? should we store only materialized list?

pakrym-oai · 2026-04-23T22:06:47Z

+    /// Turn-local environment override. This is consumed while building a turn
+    /// context and does not mutate the sticky thread environments stored on
+    /// `SessionConfiguration`.
+    pub(crate) environments: TurnEnvironmentOverride,


Should this be Option<> ? like other fields/

good call - swapped to Option, which means we can drop what was previously an enum class serving pretty much the same purpose

(codex used an enum class to be more explicit on the two modes, but Option<> is more consistent with how we're using env everywehre else)

pakrym-oai · 2026-04-23T22:10:35Z

                sub_id,
                session_configuration,
                updates.final_output_json_schema,
+                effective_environments,


doesn't feel like we should put effective_environments on turn considering we have the entire resolved list.

pakrym-oai · 2026-04-23T22:12:03Z

            let mut state = self.state.lock().await;
            match state.session_configuration.clone().apply(&updates) {
                Ok(next) => {
+                    let effective_environments =


Not sure I understand what is happening here.

In theory state.session_configuration.clone().apply(&updates) will apply updates to the config and return the new environment list as part of next. Why do we need to resolve_ref here ?

^ that's just the thread level settings - don't we still need to apply the turn level overrides?

Store the selected environment id on each resolved turn environment so turn context does not need to carry a parallel selection list. Sub-agent inheritance now projects selections from the materialized turn environments. Co-authored-by: Codex <noreply@openai.com>

Use Option<Vec<TurnEnvironmentSelection>> directly for turn-local environment overrides. None inherits sticky thread environments, while Some([]) explicitly disables environments for the turn. Co-authored-by: Codex <noreply@openai.com>

This was referenced Apr 21, 2026

Load named environments from config #18898

Draft

Wire selected environments into tools #18899

Draft

starr-openai force-pushed the starr/sticky-env-split-1-api-thread branch from ee1a5f1 to c3f1adb Compare April 21, 2026 22:11

starr-openai force-pushed the turn-environments-20260417 branch 2 times, most recently from 3fdd893 to 8dac6ca Compare April 21, 2026 23:41

starr-openai force-pushed the starr/sticky-env-split-1-api-thread branch 3 times, most recently from e04ef64 to 6edcdf2 Compare April 22, 2026 00:08

starr-openai changed the base branch from turn-environments-20260417 to starr/sticky-env-split-base-b49 April 22, 2026 00:08

starr-openai force-pushed the starr/sticky-env-split-1-api-thread branch from c8c8e86 to 47d2133 Compare April 22, 2026 01:16

github-actions Bot mentioned this pull request Apr 22, 2026

📊 AI CLI 工具社区动态日报 2026-04-22 gsscsd/big_model_radar#225

Open

starr-openai force-pushed the starr/sticky-env-split-1-api-thread branch 3 times, most recently from 25decb8 to 606663d Compare April 22, 2026 01:34

starr-openai marked this pull request as ready for review April 22, 2026 01:55

starr-openai requested a review from a team as a code owner April 22, 2026 01:55

starr-openai force-pushed the starr/sticky-env-split-1-api-thread branch 2 times, most recently from b810587 to a578544 Compare April 22, 2026 02:03

chatgpt-codex-connector Bot reviewed Apr 22, 2026

View reviewed changes

Comment thread codex-rs/core/src/session/turn_context.rs Outdated

Comment thread codex-rs/core/src/thread_manager.rs Outdated

starr-openai force-pushed the starr/sticky-env-split-1-api-thread branch from a578544 to 309cfc8 Compare April 22, 2026 02:55