Pass more params to compaction (#14247)

Pass more params to /compact. This should give us parity with the
/responses endpoint to improve caching.

I'm torn about the MCP await. Blocking will give us parity but it seems
like we explicitly don't block on MCPs. Happy either way

codex-rs/core/src/client.rs

@@ -281,6 +281,8 @@ impl ModelClient {
         &self,
         prompt: &Prompt,
         model_info: &ModelInfo,
+        effort: Option<ReasoningEffortConfig>,
+        summary: ReasoningSummaryConfig,
         session_telemetry: &SessionTelemetry,
     ) -> Result<Vec<ResponseItem>> {
         if prompt.input.is_empty() {
@@ -294,10 +296,29 @@ impl ModelClient {
                 .with_telemetry(Some(request_telemetry));
 
         let instructions = prompt.base_instructions.text.clone();
+        let input = prompt.get_formatted_input();
+        let tools = create_tools_json_for_responses_api(&prompt.tools)?;
+        let reasoning = Self::build_reasoning(model_info, effort, summary);
+        let verbosity = if model_info.support_verbosity {
+            self.state.model_verbosity.or(model_info.default_verbosity)
+        } else {
+            if self.state.model_verbosity.is_some() {
+                warn!(
+                    "model_verbosity is set but ignored as the model does not support verbosity: {}",
+                    model_info.slug
+                );
+            }
+            None
+        };
+        let text = create_text_param_for_request(verbosity, &prompt.output_schema);
         let payload = ApiCompactionInput {
             model: &model_info.slug,
-            input: &prompt.input,
+            input: &input,
             instructions: &instructions,
+            tools,
+            parallel_tool_calls: prompt.parallel_tool_calls,
+            reasoning,
+            text,
         };
 
         let mut extra_headers = self.build_subagent_headers();
@@ -375,6 +396,25 @@ impl ModelClient {
         request_telemetry
     }
 
+    fn build_reasoning(
+        model_info: &ModelInfo,
+        effort: Option<ReasoningEffortConfig>,
+        summary: ReasoningSummaryConfig,
+    ) -> Option<Reasoning> {
+        if model_info.supports_reasoning_summaries {
+            Some(Reasoning {
+                effort: effort.or(model_info.default_reasoning_level),
+                summary: if summary == ReasoningSummaryConfig::None {
+                    None
+                } else {
+                    Some(summary)
+                },
+            })
+        } else {
+            None
+        }
+    }
+
     /// Returns whether the Responses-over-WebSocket transport is active for this session.
     ///
     /// This combines provider capability and feature gating; both must be true for websocket paths

codex-rs/core/src/codex.rs

@@ -6227,7 +6227,7 @@ async fn run_sampling_request(
     }
 }
 
-async fn built_tools(
+pub(crate) async fn built_tools(
     sess: &Session,
     turn_context: &TurnContext,
     input: &[ResponseItem],

codex-rs/core/src/compact_remote.rs

@@ -1,8 +1,10 @@
+use std::collections::HashSet;
 use std::sync::Arc;
 
 use crate::Prompt;
 use crate::codex::Session;
 use crate::codex::TurnContext;
+use crate::codex::built_tools;
 use crate::compact::InitialContextInjection;
 use crate::compact::insert_initial_context_before_last_real_user_or_summary;
 use crate::context_manager::ContextManager;
@@ -19,6 +21,7 @@ use codex_protocol::items::TurnItem;
 use codex_protocol::models::BaseInstructions;
 use codex_protocol::models::ResponseItem;
 use futures::TryFutureExt;
+use tokio_util::sync::CancellationToken;
 use tracing::error;
 use tracing::info;
 
@@ -92,10 +95,20 @@ async fn run_remote_compact_task_inner_impl(
         .cloned()
         .collect();
 
+    let prompt_input = history.for_prompt(&turn_context.model_info.input_modalities);
+    let tool_router = built_tools(
+        sess.as_ref(),
+        turn_context.as_ref(),
+        &prompt_input,
+        &HashSet::new(),
+        None,
+        &CancellationToken::new(),
+    )
+    .await?;
     let prompt = Prompt {
-        input: history.for_prompt(&turn_context.model_info.input_modalities),
-        tools: vec![],
-        parallel_tool_calls: false,
+        input: prompt_input,
+        tools: tool_router.specs(),
+        parallel_tool_calls: turn_context.model_info.supports_parallel_tool_calls,
         base_instructions,
         personality: turn_context.personality,
         output_schema: None,
@@ -107,6 +120,8 @@ async fn run_remote_compact_task_inner_impl(
         .compact_conversation_history(
             &prompt,
             &turn_context.model_info,
+            turn_context.reasoning_effort,
+            turn_context.reasoning_summary,
             &turn_context.session_telemetry,
         )
         .or_else(|err| async {