fix: increase max_tokens to 8192 for R1 reasoning overhead

R1 models use 500-2000 tokens for <think> blocks before the final
response. 4096 was too tight — the model would exhaust the budget
mid-thought and never emit the JSON.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

src/claude.rs

@@ -63,7 +63,7 @@ impl ClaudeClient {
     ) -> Result<(String, Option<Usage>)> {
         let body = MessagesRequest {
             model: self.model.clone(),
-            max_tokens: 4096,
+            max_tokens: 8192,
             system: system.to_string(),
             messages: messages.to_vec(),
         };
@@ -98,9 +98,13 @@ impl ClaudeClient {
     }
 }
 
-/// Extract a JSON object from Claude's response text.
-/// Looks for the first `{` ... `}` block, handling markdown code fences.
+/// Extract a JSON object from a model response text.
+/// Handles markdown code fences and R1-style `<think>...</think>` blocks.
 pub fn extract_json(text: &str) -> Result<Value> {
+    // Strip R1-style thinking blocks before looking for JSON
+    let text = strip_think_blocks(text);
+    let text = text.as_ref();
+
     // Strip markdown fences if present
     let cleaned = text
         .replace("```json", "")
@@ -137,3 +141,25 @@ pub fn extract_json(text: &str) -> Result<Value> {
 
     serde_json::from_str(&cleaned[s..e]).context("parse extracted JSON")
 }
+
+/// Remove `<think>...</think>` blocks emitted by R1-family reasoning models.
+/// Handles nested tags and unterminated blocks (truncated responses).
+fn strip_think_blocks(text: &str) -> std::borrow::Cow<'_, str> {
+    if !text.contains("<think>") {
+        return std::borrow::Cow::Borrowed(text);
+    }
+    let mut out = String::with_capacity(text.len());
+    let mut rest = text;
+    while let Some(start) = rest.find("<think>") {
+        out.push_str(&rest[..start]);
+        rest = &rest[start + "<think>".len()..];
+        if let Some(end) = rest.find("</think>") {
+            rest = &rest[end + "</think>".len()..];
+        } else {
+            // Unterminated — discard the rest (truncated thinking block)
+            rest = "";
+        }
+    }
+    out.push_str(rest);
+    std::borrow::Cow::Owned(out)
+}