feat(engine): context-limit handoff thresholds (closes #664)

crates/tui/src/compaction.rs

@@ -25,6 +25,10 @@ pub struct CompactionConfig {
     pub message_threshold: usize,
     pub model: String,
     pub cache_summary: bool,
+    /// When `Some(true)`, the engine prefers writing a handoff file over
+    /// running LLM-based compaction when context pressure triggers.
+    /// `None` / `Some(false)` means default compaction behaviour.
+    pub prefer_handoff: Option<bool>,
 }
 
 impl Default for CompactionConfig {
@@ -40,6 +44,7 @@ impl Default for CompactionConfig {
             message_threshold: 50,
             model: DEFAULT_TEXT_MODEL.to_string(),
             cache_summary: true,
+            prefer_handoff: None,
         }
     }
 }

crates/tui/src/handoff.rs

@@ -0,0 +1,8 @@
+pub const THRESHOLDS: [(f32, &str); 3] = [
+    (0.9, "Context at 90%: stop and write handoff to .deepseek/handoff.md now"),
+    (0.8, "Context at 80%: draft handoff to .deepseek/handoff.md"),
+    (0.7, "Context at 70%: consider wrapping current sub-task"),
+];
+pub fn threshold_message(ratio: f32) -> Option<&'static str> {
+    THRESHOLDS.iter().find(|(t,_)| ratio >= *t).map(|(_,m)| *m)
+}

crates/tui/src/main.rs

@@ -30,6 +30,7 @@ mod error_taxonomy;
 mod eval;
 mod execpolicy;
 mod features;
+mod handoff;
 mod hooks;
 mod llm_client;
 mod localization;