Support typed content in contextual prompt fragments (#54776)

## Why

Contextual fragments and extension-owned world-state sections currently return strings, limiting them to text content.

## What changed

- Return `ContentItem` from fragment bodies and rendering, preserving classification and attribution. Apply text markers only to `InputText`; pass other content through unchanged.
- Allow extension prompt and world-state contributions to carry typed content, and preserve media in context-window hints.
- Hash non-text world-state fragments by their serialized content and role while retaining the existing text hashing behavior.
- Add shared content-to-tool-output conversion that preserves image references, image detail, and audio.

## Testing

Add coverage for typed world-state hashing and preservation of image content and attribution. Expand code-mode output tests to cover input text, output text, file-backed and inline images, and audio.

GitOrigin-RevId: 2ed00d466a3f169b151202f3eef71e667ea64be1

codex-rs/core/src/context_manager/history_tests.rs

@@ -27,6 +27,7 @@ use codex_protocol::models::LocalShellExecAction;
 use codex_protocol::models::LocalShellStatus;
 use codex_protocol::models::ReasoningItemContent;
 use codex_protocol::models::ReasoningItemReasoningSummary;
+use codex_protocol::models::input_text;
 use codex_protocol::openai_models::InputModality;
 use codex_protocol::openai_models::ReasoningEffort;
 use codex_protocol::openai_models::default_input_modalities;
@@ -37,6 +38,7 @@ use codex_protocol::protocol::InterAgentCommunication;
 use codex_protocol::protocol::PLUGINS_INSTRUCTIONS_OPEN_TAG;
 use codex_protocol::protocol::SandboxPolicy;
 use codex_protocol::protocol::TurnContextItem;
+use codex_protocol_test_support::ContentItemTestExt;
 use codex_utils_absolute_path::AbsolutePathBuf;
 use codex_utils_output_truncation::TruncationPolicy;
 use codex_utils_output_truncation::truncate_text;
@@ -601,7 +603,7 @@ fn world_state_transitions_persist_changed_state_and_skip_unchanged_state() {
                 return (None, None);
             }
             (Some(snapshot), Some(RenderedWorldStateFragment::new(
-                "developer", ("<test>", "</test>"), value,
+                "developer", ("<test>", "</test>"), input_text(value.to_string()),
             )))
         }));
         state
@@ -615,7 +617,7 @@ fn world_state_transitions_persist_changed_state_and_skip_unchanged_state() {
     let WorldStateUpdateContent::Fragment(fragment) = &fragments[0].content else {
         panic!("expected a text fragment");
     };
-    assert_eq!(fragment.body(), "after");
+    assert_eq!(fragment.body().expect_text(), "after");
     let patch = patch.expect("changed snapshot must be persisted");
     assert!(!patch.full);
     let mut replayed = WorldStateSnapshot::from(&full.state);
@@ -646,7 +648,7 @@ fn world_state_reconciles_matching_legacy_history_once() {
         fragments
             .into_iter()
             .map(|update| match update.content {
-                WorldStateUpdateContent::Fragment(fragment) => fragment.body(),
+                WorldStateUpdateContent::Fragment(fragment) => fragment.body().expect_text(),
                 WorldStateUpdateContent::Item(item) =>
                     panic!("expected a text fragment, got {item:?}"),
             })
@@ -1943,7 +1945,9 @@ fn drop_last_n_user_turns_trims_context_updates_above_rolled_back_turn() {
 #[test]
 fn drop_last_n_user_turns_preserves_annotations_for_surviving_developer_fragments() {
     let turn_id = "rolled-back-turn";
-    let model_switch = ModelSwitchInstructions::new("switched model instructions").render();
+    let model_switch = ModelSwitchInstructions::new("switched model instructions")
+        .render()
+        .expect_text();
     let developer_message = ResponseItem::Message {
         status: None,
         encrypted_content: None,

codex-rs/core/src/context_manager/normalize.rs

@@ -372,11 +372,8 @@ pub(crate) fn strip_images_when_unsupported(
                     for content_item in content_items.iter() {
                         match content_item {
                             FunctionCallOutputContentItem::InputImage { .. } => {
-                                normalized_content_items.push(
-                                    FunctionCallOutputContentItem::InputText {
-                                        text: UnsupportedMedia::IMAGE.render(),
-                                    },
-                                );
+                                normalized_content_items
+                                    .push(UnsupportedMedia::IMAGE.render().into());
                             }
                             _ => normalized_content_items.push(content_item.clone()),
                         }
@@ -423,9 +420,7 @@ pub(crate) fn strip_audio_when_unsupported(
                             content_item,
                             FunctionCallOutputContentItem::InputAudio { .. }
                         ) {
-                            *content_item = FunctionCallOutputContentItem::InputText {
-                                text: UnsupportedMedia::AUDIO.render(),
-                            };
+                            *content_item = UnsupportedMedia::AUDIO.render().into();
                         }
                     }
                 }

codex-rs/core/src/context_manager/updates_tests.rs

@@ -5,6 +5,7 @@ use crate::context::ContextualUserFragment;
 use codex_protocol::models::ContentItem;
 use codex_protocol::models::ContentItemKind;
 use codex_protocol::models::InternalChatMessageMetadataPassthrough;
+use codex_protocol::models::input_text;
 use pretty_assertions::assert_eq;
 use serde_json::json;
 
@@ -31,8 +32,8 @@ impl ContextualUserFragment for TestFragment {
         ("", "")
     }
 
-    fn body(&self) -> String {
-        self.text.to_string()
+    fn body(&self) -> ContentItem {
+        input_text(self.text.to_string())
     }
 }