Use the image generation extension by default

codex-rs/app-server/tests/suite/v2/imagegen_extension.rs

@@ -588,7 +588,6 @@ model_provider = "openai-custom"
 chatgpt_base_url = "{server_uri}"
 
 [features]
-imagegenext = true
 {code_mode_only}
 
 [model_providers.openai-custom]

codex-rs/core-api/Cargo.toml

@@ -21,6 +21,7 @@ codex-config = { workspace = true }
 codex-core = { workspace = true }
 codex-extension-api = { workspace = true }
 codex-home = { workspace = true }
+codex-image-generation-extension = { workspace = true }
 codex-exec-server = { workspace = true }
 codex-features = { workspace = true }
 codex-login = { workspace = true }

codex-rs/core-api/src/lib.rs

@@ -61,6 +61,7 @@ pub use codex_exec_server::NoiseChannelIdentity;
 pub use codex_exec_server::NoiseChannelPublicKey;
 pub use codex_exec_server::NoiseRendezvousConnectBundle;
 pub use codex_exec_server::NoiseRendezvousConnectProvider;
+pub use codex_extension_api::ExtensionRegistryBuilder;
 pub use codex_extension_api::LoadUserInstructionsFuture;
 pub use codex_extension_api::LoadedUserInstructions;
 pub use codex_extension_api::UserInstructions;
@@ -69,6 +70,7 @@ pub use codex_extension_api::empty_extension_registry;
 pub use codex_features::Feature;
 pub use codex_features::Features;
 pub use codex_home::CodexHomeUserInstructionsProvider;
+pub use codex_image_generation_extension::install as install_image_generation_extension;
 pub use codex_login::AuthHeaders;
 pub use codex_login::AuthManager;
 pub use codex_login::CodexAuth;

codex-rs/core/src/context/image_generation_instructions.rs

@@ -1,4 +1,3 @@
-use super::ContextualUserFragment;
 use std::fmt::Display;
 
 /// Maximum size of the extension's model-facing generated-image path hint.
@@ -21,36 +20,3 @@ fn image_generation_hint(
         "Generated images are saved to {image_output_dir} as {image_output_path} by default.\nIf you need to use a generated image at another path, copy it and leave the original in place unless the user explicitly asks you to delete it."
     )
 }
-
-#[derive(Debug, Clone, PartialEq)]
-pub(crate) struct ImageGenerationInstructions {
-    image_output_dir: String,
-    image_output_path: String,
-}
-
-impl ImageGenerationInstructions {
-    pub(crate) fn new(image_output_dir: impl Display, image_output_path: impl Display) -> Self {
-        Self {
-            image_output_dir: image_output_dir.to_string(),
-            image_output_path: image_output_path.to_string(),
-        }
-    }
-}
-
-impl ContextualUserFragment for ImageGenerationInstructions {
-    fn role(&self) -> &'static str {
-        "developer"
-    }
-
-    fn markers(&self) -> (&'static str, &'static str) {
-        Self::type_markers()
-    }
-
-    fn type_markers() -> (&'static str, &'static str) {
-        ("", "")
-    }
-
-    fn body(&self) -> String {
-        image_generation_hint(&self.image_output_dir, &self.image_output_path)
-    }
-}

codex-rs/core/src/context/mod.rs

@@ -50,7 +50,6 @@ pub(crate) use contextual_user_message::parse_visible_hook_prompt_message;
 pub(crate) use current_time_reminder::CurrentTimeReminder;
 pub(crate) use guardian_followup_review_reminder::GuardianFollowupReviewReminder;
 pub(crate) use hook_additional_context::HookAdditionalContext;
-pub(crate) use image_generation_instructions::ImageGenerationInstructions;
 pub use image_generation_instructions::extension_image_generation_output_hint;
 pub(crate) use inter_agent_completion_message::InterAgentCompletionMessage;
 pub use internal_model_context::InternalContextSource;

codex-rs/core/src/session/tests.rs

@@ -8512,49 +8512,6 @@ async fn build_initial_context_omits_multi_agent_v2_usage_hints_when_hint_is_emp
     );
 }
 
-#[tokio::test]
-async fn build_initial_context_omits_default_image_save_location_with_image_history() {
-    let (session, turn_context) = make_session_and_context().await;
-    session
-        .replace_history(
-            vec![ResponseItem::ImageGenerationCall {
-                id: Some("ig-test".to_string()),
-                status: "completed".to_string(),
-                revised_prompt: Some("a tiny blue square".to_string()),
-                result: "Zm9v".to_string(),
-                internal_chat_message_metadata_passthrough: None,
-            }],
-            /*reference_context_item*/ None,
-        )
-        .await;
-    let turn_context = Arc::new(turn_context);
-
-    let initial_context = build_initial_context(&session, &turn_context).await;
-    let developer_texts = developer_input_texts(&initial_context);
-    assert!(
-        !developer_texts
-            .iter()
-            .any(|text| text.contains("Generated images are saved to")),
-        "expected initial context to omit image save instructions even with image history, got {developer_texts:?}"
-    );
-}
-
-#[tokio::test]
-async fn build_initial_context_omits_default_image_save_location_without_image_history() {
-    let (session, turn_context) = make_session_and_context().await;
-    let turn_context = Arc::new(turn_context);
-
-    let initial_context = build_initial_context(&session, &turn_context).await;
-    let developer_texts = developer_input_texts(&initial_context);
-
-    assert!(
-        !developer_texts
-            .iter()
-            .any(|text| text.contains("Generated images are saved to")),
-        "expected initial context to omit image save instructions without image history, got {developer_texts:?}"
-    );
-}
-
 #[tokio::test]
 async fn build_initial_context_trims_skill_metadata_from_context_window_budget() {
     let (session, mut turn_context) = make_session_and_context().await;
@@ -8762,102 +8719,6 @@ async fn build_initial_context_emits_thread_start_skill_warning_on_repeated_buil
     ));
 }
 
-#[tokio::test]
-async fn handle_output_item_done_records_image_save_history_message() {
-    let (session, turn_context) = make_session_and_context().await;
-    let session = Arc::new(session);
-    let turn_context = Arc::new(turn_context);
-    let call_id = "ig_history_records_message";
-    let expected_saved_path = crate::stream_events_utils::image_generation_artifact_path(
-        &turn_context.config.codex_home,
-        &session.thread_id.to_string(),
-        call_id,
-    );
-    let _ = std::fs::remove_file(&expected_saved_path);
-    let item = ResponseItem::ImageGenerationCall {
-        id: Some(call_id.to_string()),
-        status: "completed".to_string(),
-        revised_prompt: Some("a tiny blue square".to_string()),
-        result: "Zm9v".to_string(),
-        internal_chat_message_metadata_passthrough: None,
-    };
-
-    let mut ctx = HandleOutputCtx {
-        sess: Arc::clone(&session),
-        turn_context: Arc::clone(&turn_context),
-        turn_store: Arc::new(codex_extension_api::ExtensionData::new(
-            turn_context.sub_id.clone(),
-        )),
-        tool_runtime: test_tool_runtime(Arc::clone(&session), Arc::clone(&turn_context)),
-        cancellation_token: CancellationToken::new(),
-    };
-    handle_output_item_done(&mut ctx, item.clone(), /*previously_active_item*/ None)
-        .await
-        .expect("image generation item should succeed");
-
-    let history = session.clone_history().await;
-    let image_output_path = crate::stream_events_utils::image_generation_artifact_path(
-        &turn_context.config.codex_home,
-        &session.thread_id.to_string(),
-        "<image_id>",
-    );
-    let image_output_dir = image_output_path
-        .parent()
-        .expect("generated image path should have a parent");
-    let image_message: ResponseItem = crate::context::ContextualUserFragment::into(
-        crate::context::ImageGenerationInstructions::new(
-            image_output_dir.display(),
-            image_output_path.display(),
-        ),
-    );
-    let expected = vec![image_message, item];
-    assert_eq!(strip_metadata_from_items(history.raw_items()), expected);
-    assert_eq!(
-        std::fs::read(&expected_saved_path).expect("saved file"),
-        b"foo"
-    );
-    let _ = std::fs::remove_file(&expected_saved_path);
-}
-
-#[tokio::test]
-async fn handle_output_item_done_skips_image_save_message_when_save_fails() {
-    let (session, turn_context) = make_session_and_context().await;
-    let session = Arc::new(session);
-    let turn_context = Arc::new(turn_context);
-    let call_id = "ig_history_no_message";
-    let expected_saved_path = crate::stream_events_utils::image_generation_artifact_path(
-        &turn_context.config.codex_home,
-        &session.thread_id.to_string(),
-        call_id,
-    );
-    let _ = std::fs::remove_file(&expected_saved_path);
-    let item = ResponseItem::ImageGenerationCall {
-        id: Some(call_id.to_string()),
-        status: "completed".to_string(),
-        revised_prompt: Some("broken payload".to_string()),
-        result: "_-8".to_string(),
-        internal_chat_message_metadata_passthrough: None,
-    };
-
-    let mut ctx = HandleOutputCtx {
-        sess: Arc::clone(&session),
-        turn_context: Arc::clone(&turn_context),
-        turn_store: Arc::new(codex_extension_api::ExtensionData::new(
-            turn_context.sub_id.clone(),
-        )),
-        tool_runtime: test_tool_runtime(Arc::clone(&session), Arc::clone(&turn_context)),
-        cancellation_token: CancellationToken::new(),
-    };
-    handle_output_item_done(&mut ctx, item.clone(), /*previously_active_item*/ None)
-        .await
-        .expect("image generation item should still complete");
-
-    let history = session.clone_history().await;
-    let expected = vec![item];
-    assert_eq!(strip_metadata_from_items(history.raw_items()), expected);
-    assert!(!expected_saved_path.exists());
-}
-
 #[tokio::test]
 async fn build_initial_context_uses_previous_turn_settings_for_realtime_end() {
     let (session, turn_context) = make_session_and_context().await;

codex-rs/core/src/session/turn.rs

@@ -1853,7 +1853,6 @@ async fn handle_assistant_item_done_in_plan_mode(
         let mut finalized_facts = None;
         if let Some(finalized_turn_item) = finalize_non_tool_response_item(
             sess,
-            turn_context,
             TurnItemContributorPolicy::Run(turn_store),
             item,
             /*plan_mode*/ true,
@@ -2141,7 +2140,6 @@ async fn try_run_sampling_request(
                 }
                 if let Some(turn_item) = handle_non_tool_response_item(
                     sess.as_ref(),
-                    turn_context.as_ref(),
                     TurnItemContributorPolicy::Skip,
                     &item,
                     plan_mode,