diff --git a/src/agent/context.rs b/src/agent/context.rs
index d61136f..a42beeb 100644
--- a/src/agent/context.rs
+++ b/src/agent/context.rs
@@ -900,7 +900,43 @@ impl ResponseParser {
     }
 
     pub fn finish(mut self, ctx: &mut ContextState) {
-        if !self.buf.is_empty() {
+        // Salvage any in-flight tag accumulators if the stream ended
+        // before the close tag arrived (max_tokens, premature EOS,
+        // server-side cancel). Without this, an unterminated
+        // <think>...</think> drops all of self.think_buf and only the
+        // trailing rolling window in self.buf survives — observed as
+        // "responses cut off, only the last ~8 characters come
+        // through" because drain_safe keeps `close_tag.len()` bytes
+        // (8 for `</think>`) at the tail of buf.
+        if self.in_think {
+            if !self.buf.is_empty() {
+                self.think_buf.push_str(&std::mem::take(&mut self.buf));
+            }
+            let text = std::mem::take(&mut self.think_buf).trim().to_string();
+            if !text.is_empty() {
+                self.push_child(ctx, AstNode::thinking(text));
+            }
+            self.in_think = false;
+        } else if self.in_tool_call {
+            if !self.buf.is_empty() {
+                self.tool_call_buf.push_str(&std::mem::take(&mut self.buf));
+            }
+            let body = std::mem::take(&mut self.tool_call_buf);
+            match parse_tool_call_body(&body) {
+                Some((name, args)) => {
+                    self.flush_content(ctx);
+                    self.push_child(ctx, AstNode::tool_call(&name, &args));
+                }
+                None => {
+                    // Body's likely incomplete (no `</tool_call>` ever
+                    // arrived). Wrap as content with the open tag so the
+                    // model can see its own truncated attempt next turn
+                    // rather than losing it silently.
+                    self.content_parts.push(format!("<tool_call>\n{}", body));
+                }
+            }
+            self.in_tool_call = false;
+        } else if !self.buf.is_empty() {
             self.content_parts.push(std::mem::take(&mut self.buf));
         }
         self.flush_content(ctx);