Clamp auto-compact limit to context window (#11516)

- Clamp auto-compaction to the minimum of configured limit and 90% of
context window
- Add an e2e compact test for clamped behavior
- Update remote compact tests to account for earlier auto-compaction in
setup turns

codex-rs/core/tests/suite/compact.rs

@@ -2553,6 +2553,64 @@ async fn auto_compact_triggers_after_function_call_over_95_percent_usage() {
     );
 }
 
+#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+async fn auto_compact_clamps_config_limit_to_context_window() {
+    skip_if_no_network!();
+
+    let server = start_mock_server().await;
+
+    let context_window = 100;
+    let config_limit = 200;
+    let over_limit_tokens = context_window * 90 / 100 + 1;
+
+    let first_turn = sse(vec![
+        ev_assistant_message("m1", FIRST_REPLY),
+        ev_completed_with_tokens("r1", over_limit_tokens),
+    ]);
+    let auto_summary_payload = auto_summary(AUTO_SUMMARY_TEXT);
+    let auto_compact_turn = sse(vec![
+        ev_assistant_message("m2", &auto_summary_payload),
+        ev_completed_with_tokens("r2", 10),
+    ]);
+    let post_auto_compact_turn = sse(vec![ev_completed_with_tokens("r3", 10)]);
+
+    let first_turn_mock = mount_sse_once(&server, first_turn).await;
+    let auto_compact_mock = mount_sse_once(&server, auto_compact_turn).await;
+    mount_sse_once(&server, post_auto_compact_turn).await;
+
+    let model_provider = non_openai_model_provider(&server);
+    let mut builder = test_codex().with_config(move |config| {
+        config.model_provider = model_provider;
+        set_test_compact_prompt(config);
+        config.model_context_window = Some(context_window);
+        config.model_auto_compact_token_limit = Some(config_limit);
+    });
+    let codex = builder.build(&server).await.unwrap();
+
+    codex.submit_turn("OVER_LIMIT_TURN").await.unwrap();
+    codex.submit_turn("FOLLOW_UP_AFTER_CLAMP").await.unwrap();
+
+    assert!(
+        first_turn_mock.single_request().input().iter().any(|item| {
+            item.get("type").and_then(|value| value.as_str()) == Some("message")
+                && item
+                    .get("content")
+                    .and_then(|content| content.as_array())
+                    .and_then(|entries| entries.first())
+                    .and_then(|entry| entry.get("text"))
+                    .and_then(|value| value.as_str())
+                    == Some("OVER_LIMIT_TURN")
+        }),
+        "first request should contain the over-limit user input"
+    );
+
+    let auto_compact_body = auto_compact_mock.single_request().body_json().to_string();
+    assert!(
+        body_contains_text(&auto_compact_body, SUMMARIZATION_PROMPT),
+        "auto compact should run with the summarization prompt when config limit exceeds context"
+    );
+}
+
 #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
 async fn auto_compact_counts_encrypted_reasoning_before_last_user() {
     skip_if_no_network!();