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/ext/extension-api/src/contributors/prompt.rs

@@ -11,7 +11,7 @@ use codex_protocol::models::ContentItemMetadata;
 pub enum PromptSlot {
     DeveloperPolicy,
     DeveloperCapabilities,
-    /// Text inside the context-window message, supplied by `contribute_thread_context`.
+    /// Content inside the context-window message, supplied by `contribute_thread_context`.
     ContextWindow,
 }
 
@@ -22,16 +22,12 @@ pub struct PromptFragment {
 }
 
 impl PromptFragment {
-    /// Places a contextual fragment in the prompt, preserving its text, kind, and source.
+    /// Places a contextual fragment in the prompt, preserving its content, kind, and source.
     /// The slot determines placement within the developer message, regardless of its original role.
     pub fn from_fragment(slot: PromptSlot, fragment: &impl ContextualUserFragment) -> Self {
         Self {
             slot,
-            content: AnnotatedContent::text(
-                fragment.render(),
-                fragment.content_kind(),
-                fragment.content_metadata(),
-            ),
+            content: fragment.render_fragment().into_parts().1,
         }
     }
 
@@ -75,15 +71,12 @@ impl PromptFragment {
         self.slot
     }
 
-    /// Returns the model-visible text.
-    pub fn text(&self) -> &str {
-        let ContentItem::InputText { text } = self.content.content() else {
-            unreachable!("prompt fragments contain text");
-        };
-        text
+    /// Returns the model-visible content.
+    pub fn content(&self) -> &ContentItem {
+        self.content.content()
     }
 
-    /// Returns the producer-owned classification of the model-visible text.
+    /// Returns the producer-owned classification of the model-visible content.
     pub fn content_kind(&self) -> &ContentItemKind {
         self.content.kind()
     }

codex-rs/ext/extension-api/src/contributors/world_state.rs

@@ -4,6 +4,7 @@ use codex_exec_server_protocol::ExecutorCapabilityDiscoverySnapshot;
 use codex_exec_server_protocol::ExecutorCapabilityDiscoveryV2Snapshot;
 use codex_protocol::ThreadId;
 use codex_protocol::capabilities::SelectedCapabilityRoot;
+use codex_protocol::models::ContentItem;
 use codex_protocol::models::ContentItemMetadata;
 use codex_protocol::openai_models::ModelInfo;
 use codex_protocol::protocol::TurnEnvironmentSelection;
@@ -45,12 +46,15 @@ pub enum PreviousWorldStateSection<'a> {
     Known(&'a Value),
 }
 
-/// Plain model-visible data rendered by an extension-owned World State section.
-#[derive(Clone, Debug, PartialEq, Eq)]
+/// Model-visible content rendered by an extension-owned World State section.
+/// Markers apply only to `InputText` bodies.
+/// The producer owns size bounds for the content; rendering does not truncate it.
+/// Non-text user-role bodies are currently classified as ordinary user media.
+#[derive(Clone, Debug, PartialEq)]
 pub struct RenderedWorldStateFragment {
     role: &'static str,
     markers: (&'static str, &'static str),
-    body: String,
+    body: ContentItem,
     metadata: ContentItemMetadata,
 }
 
@@ -58,12 +62,12 @@ impl RenderedWorldStateFragment {
     pub fn new(
         role: &'static str,
         markers: (&'static str, &'static str),
-        body: impl Into<String>,
+        body: ContentItem,
     ) -> Self {
         Self {
             role,
             markers,
-            body: body.into(),
+            body,
             metadata: ContentItemMetadata::harness(),
         }
     }
@@ -86,7 +90,7 @@ impl RenderedWorldStateFragment {
         self.markers
     }
 
-    pub fn body(&self) -> &str {
+    pub fn body(&self) -> &ContentItem {
         &self.body
     }
 }

codex-rs/ext/extension-api/src/lib.rs

@@ -22,8 +22,10 @@ pub use capabilities::ResponseItemInjector;
 pub use codex_context_fragments::ContextualUserFragment;
 pub use codex_mcp::McpProtocolMode;
 pub use codex_mcp::ToolInfo as McpToolInfo;
+pub use codex_protocol::models::ContentItem;
 pub use codex_protocol::models::ContentItemKind;
 pub use codex_protocol::models::ResponseItem;
+pub use codex_protocol::models::input_text;
 pub use codex_protocol::security_risk::SecurityRiskScore;
 pub use codex_tools::ConversationHistory;
 pub use codex_tools::ExtensionTurnItem;