From 0383166f4ecd42b4d76f298c173080c14ea1159b Mon Sep 17 00:00:00 2001
From: Ryan Ward <ryan@ryserve.net>
Date: Tue, 20 Oct 2026 18:27:34 +0000
Subject: [PATCH] Use the active model context limit for provider compaction

---
 app/src/ai/blocklist/controller.rs            |  5 +-
 .../ai/runtime/provider_run_coordinator.rs    |  9 ++++
 .../runtime/provider_run_coordinator_tests.rs | 50 +++++++++++++++++++
 3 files changed, 62 insertions(+), 2 deletions(-)

diff --git a/app/src/ai/blocklist/controller.rs b/app/src/ai/blocklist/controller.rs
index 7a23ffd3..bfafeb5a 100644
--- a/app/src/ai/blocklist/controller.rs
+++ b/app/src/ai/blocklist/controller.rs
@@ -6935,8 +6935,9 @@ impl BlocklistAIController {
         }
 
         let active_context_limit = active_run
-            .response_config
-            .max_context_tokens
+            .coordinator
+            .current_context_limit()
+            .or(active_run.response_config.max_context_tokens)
             .unwrap_or_else(|| {
                 let preferences = LLMPreferences::as_ref(ctx);
                 configured_llm_context_limit(
diff --git a/app/src/ai/runtime/provider_run_coordinator.rs b/app/src/ai/runtime/provider_run_coordinator.rs
index 435b4ac4..391e2138 100644
--- a/app/src/ai/runtime/provider_run_coordinator.rs
+++ b/app/src/ai/runtime/provider_run_coordinator.rs
@@ -251,6 +251,15 @@ impl ProviderRunCoordinator {
         self.profiles.get(profile).map(|profile| &profile.request)
     }
 
+    /// The pending model profile owns the context budget, including after a
+    /// preflight overflow has returned the run to its compaction boundary.
+    pub(crate) fn current_context_limit(&self) -> Option<u32> {
+        self.profiles
+            .get(self.run.profile().as_str())
+            .and_then(|profile| profile.max_context_tokens)
+            .or(self.max_context_tokens)
+    }
+
     /// Includes the current transcript, fixed request context, tools, and output reservation.
     /// Read at the model boundary so newly committed tool output is included.
     pub(crate) fn estimated_context_tokens(&self) -> Option<(u32, u32)> {
diff --git a/app/src/ai/runtime/provider_run_coordinator_tests.rs b/app/src/ai/runtime/provider_run_coordinator_tests.rs
index 88003166..fa685f81 100644
--- a/app/src/ai/runtime/provider_run_coordinator_tests.rs
+++ b/app/src/ai/runtime/provider_run_coordinator_tests.rs
@@ -1558,6 +1558,56 @@ fn context_estimate_tracks_live_transcript_and_reserves_fixed_context() {
     assert_eq!(fixed_before, fixed_after);
 }
 
+#[tokio::test]
+async fn compaction_boundary_keeps_the_monitor_context_limit() {
+    let base_runtime = Arc::new(ScriptedRuntime::new(vec![]));
+    let cli_runtime = Arc::new(ScriptedRuntime::new(vec![]));
+    let run = ProviderRun::new(
+        "monitor-compaction-budget",
+        vec![ConversationMessage {
+            role: MessageRole::User,
+            content: MessageContent::Text("history".repeat(5_000)),
+        }],
+        CLI_MONITOR_PROVIDER_PROFILE,
+        ProviderRunLimits::default(),
+    );
+    let mut profiles = BTreeMap::new();
+    profiles.insert(
+        BASE_PROVIDER_PROFILE.to_string(),
+        ProviderRunProfile::new(base_runtime.clone(), request()),
+    );
+    profiles.insert(
+        CLI_MONITOR_PROVIDER_PROFILE.to_string(),
+        ProviderRunProfile::new(cli_runtime.clone(), request())
+            .with_max_context_tokens(Some(8_000)),
+    );
+    let mut coordinator = ProviderRunCoordinator::new(run, profiles).unwrap();
+    coordinator.set_max_context_tokens(Some(100_000));
+    let (_sender, control) = turn_control();
+
+    let block = coordinator
+        .drive_until_blocked(control, |_| Ok(()))
+        .await
+        .unwrap();
+
+    assert_eq!(block, ProviderRunBlock::ContextWindowExceeded);
+    assert_eq!(coordinator.run().state(), &ProviderRunState::ReadyToCallModel);
+    assert_eq!(coordinator.current_context_limit(), Some(8_000));
+    assert!(base_runtime.requests().is_empty());
+    assert!(cli_runtime.requests().is_empty());
+
+    let work_id = coordinator.run().ready_work_id().unwrap();
+    coordinator
+        .run_mut()
+        .continue_ready_with_observation(
+            &work_id,
+            MessageContent::Text("Command finished".to_string()),
+            BASE_PROVIDER_PROFILE,
+        )
+        .unwrap();
+    assert_eq!(coordinator.current_context_limit(), Some(100_000));
+}
+
 #[tokio::test]
 async fn monitor_request_uses_its_profile_context_limit() {
     let base_runtime = Arc::new(ScriptedRuntime::new(vec![answer_turn()]));