fix: FINAL() callable in REPL; parser ignores FINAL/FINAL_VAR inside code fences

rlm/environments/base_env.py

@@ -16,6 +16,7 @@
         "llm_query_batched",
         "rlm_query",
         "rlm_query_batched",
+        "FINAL",
         "FINAL_VAR",
         "SHOW_VARS",
         "context",

rlm/environments/local_repl.py

@@ -188,6 +188,7 @@ def setup(self):
         self._last_final_answer: str | None = None
 
         # Add helper functions
+        self.globals["FINAL"] = self._final
         self.globals["FINAL_VAR"] = self._final_var
         self.globals["SHOW_VARS"] = self._show_vars
         self.globals["llm_query"] = self._llm_query
@@ -205,6 +206,12 @@ def setup(self):
                 # For non-callable values (constants, data), add to locals
                 self.locals[name] = value
 
+    def _final(self, value: Any) -> str:
+        """Store and return a direct final value (compat with models calling FINAL in REPL)."""
+        answer = str(value)
+        self._last_final_answer = answer
+        return answer
+
     def _final_var(self, variable_name: str | Any) -> str:
         """Return the value of a variable as a final answer for the main model, or stringify a direct value."""
         if not isinstance(variable_name, str):
@@ -468,6 +475,8 @@ def _restore_scaffold(self) -> None:
                 self.globals["rlm_query"] = self._rlm_query
             elif name == "rlm_query_batched":
                 self.globals["rlm_query_batched"] = self._rlm_query_batched
+            elif name == "FINAL":
+                self.globals["FINAL"] = self._final
             elif name == "FINAL_VAR":
                 self.globals["FINAL_VAR"] = self._final_var
             elif name == "SHOW_VARS":

rlm/utils/parsing.py

@@ -40,9 +40,13 @@ def find_final_answer(text: str, environment: "BaseEnv | None" = None) -> str |
     Returns:
         The final answer string, or None if no final answer pattern is found
     """
+    # Remove fenced code blocks first so FINAL()/FINAL_VAR() inside ```repl``` code
+    # does not get parsed as a terminal answer.
+    text_no_code = re.sub(r"```[\s\S]*?```", "", text)
+
     # Check for FINAL_VAR pattern first - must be at start of line
     final_var_pattern = r"^\s*FINAL_VAR\((.*?)\)"
-    match = re.search(final_var_pattern, text, re.MULTILINE | re.DOTALL)
+    match = re.search(final_var_pattern, text_no_code, re.MULTILINE | re.DOTALL)
     if match:
         variable_name = match.group(1).strip().strip('"').strip("'")
         if environment is not None:
@@ -63,7 +67,7 @@ def find_final_answer(text: str, environment: "BaseEnv | None" = None) -> str |
     # Check for FINAL pattern - must be at start of line
     # Use greedy matching to capture content with nested parentheses
     final_pattern = r"^\s*FINAL\((.*)\)\s*$"
-    match = re.search(final_pattern, text, re.MULTILINE | re.DOTALL)
+    match = re.search(final_pattern, text_no_code, re.MULTILINE | re.DOTALL)
     if match:
         return match.group(1).strip()
 

tests/test_parsing.py

@@ -278,6 +278,47 @@ def test_final_var_variable_not_found_returns_none(self):
         result = find_final_answer("FINAL_VAR(missing)", environment=mock_env)
         assert result is None
 
+    def test_final_inside_repl_code_block_not_parsed_as_terminal(self):
+        """FINAL() written inside a ```repl``` block must NOT be treated as the final answer.
+
+        This was the root-cause bug: models often write FINAL(x) or FINAL_VAR(x) inside a
+        repl block as code. The parser must ignore those and only look at prose text.
+        """
+        text = "I will now finalise.\n```repl\nFINAL(final_answer)\n```"
+        result = find_final_answer(text)
+        assert result is None, (
+            "FINAL() inside a repl block should not trigger termination"
+        )
+
+    def test_final_var_inside_repl_code_block_not_parsed_as_terminal(self):
+        """FINAL_VAR() inside a repl block must NOT be treated as the final answer."""
+        text = "Computing now.\n```repl\nFINAL_VAR(result)\n```"
+        mock_env = Mock()
+        mock_env.execute_code.return_value = REPLResult(stdout="55", stderr="", locals={})
+        result = find_final_answer(text, environment=mock_env)
+        assert result is None, (
+            "FINAL_VAR() inside a repl block should not trigger termination"
+        )
+        mock_env.execute_code.assert_not_called()
+
+    def test_final_in_prose_still_works_alongside_repl_block(self):
+        """FINAL() in prose (outside code fences) must still be picked up correctly."""
+        text = "```repl\nx = 1\n```\nFINAL(55)"
+        result = find_final_answer(text)
+        assert result == "55"
+
+    def test_final_callable_in_repl_environment(self):
+        """FINAL() should be callable inside the REPL and store the answer in _last_final_answer."""
+        env = LocalREPL()
+        try:
+            env.execute_code("answer = 55")
+            result = env.execute_code("FINAL(answer)")
+            assert result.final_answer == "55", (
+                f"Expected final_answer='55', got '{result.final_answer}'"
+            )
+        finally:
+            env.cleanup()
+
 
 class TestFormatExecutionResult:
     """Tests for format_execution_result function."""