fix(reasoning): detect bare READY markers and honor the last decision

lib/crewai/src/crewai/utilities/reasoning_handler.py

@@ -4,6 +4,7 @@
 
 import json
 import logging
+import re
 from typing import TYPE_CHECKING, Any, Final, Literal, cast
 
 from pydantic import BaseModel, Field
@@ -26,6 +27,49 @@
     from crewai.agent.planning_config import PlanningConfig
     from crewai.task import Task
 
+# Full instructional phrase still emitted by the default planning prompts.
+_READY_FULL_PHRASE: Final[str] = "READY: I am ready to execute the task."
+# Models (especially local/Ollama) often conclude with a bare READY / NOT READY
+# marker at the start of a line. Anchor at line start to avoid mid-sentence hits
+# like "I am ready to begin researching".
+_READY_LINE_RE: Final[re.Pattern[str]] = re.compile(
+    r"^\s*(NOT\s+)?READY\b",
+    re.IGNORECASE,
+)
+
+
+def response_indicates_ready(response: str) -> bool:
+    """Return whether a planning response concludes the agent is READY.
+
+    Accepts the full instructional phrase and short-form markers such as a
+    trailing ``READY`` / ``READY.`` line. ``NOT READY`` always wins over a
+    bare ``READY`` substring match (e.g. inside ``NOT READY``).
+
+    When multiple markers appear (common during plan refinement), the **last**
+    explicit READY / NOT READY line is authoritative.
+
+    Args:
+        response: Raw LLM planning response text.
+
+    Returns:
+        True if the response indicates readiness to execute the task.
+    """
+    if not response:
+        return False
+
+    last_decision: bool | None = None
+    for line in response.splitlines():
+        match = _READY_LINE_RE.match(line)
+        if match is None:
+            continue
+        # Group 1 is the optional "NOT " prefix.
+        last_decision = match.group(1) is None
+
+    if last_decision is not None:
+        return last_decision
+
+    return _READY_FULL_PHRASE in response
+
 
 class ReasoningPlan(BaseModel):
     """Model representing a reasoning plan for a task."""
@@ -409,7 +453,7 @@ def _create_reasoning_plan(
             return (
                 response_str,
                 [],
-                "READY: I am ready to execute the task." in response_str,
+                response_indicates_ready(response_str),
             )
 
         except Exception as e:
@@ -433,7 +477,7 @@ def _create_reasoning_plan(
                 return (
                     fallback_str,
                     [],
-                    "READY: I am ready to execute the task." in fallback_str,
+                    response_indicates_ready(fallback_str),
                 )
             except Exception as inner_e:
                 self.logger.error(f"Error during fallback text parsing: {inner_e!s}")
@@ -593,7 +637,7 @@ def _parse_planning_response(response: str) -> tuple[str, bool]:
             return "No plan was generated.", False
 
         plan = response
-        ready = "READY: I am ready to execute the task." in response
+        ready = response_indicates_ready(response)
 
         return plan, ready
 

lib/crewai/tests/utilities/test_response_indicates_ready.py

@@ -0,0 +1,104 @@
+"""Tests for planning READY marker detection (#6204)."""
+
+import pytest
+
+from crewai.utilities.reasoning_handler import (
+    AgentReasoning,
+    response_indicates_ready,
+)
+
+
+class TestResponseIndicatesReady:
+    """Unit tests for response_indicates_ready()."""
+
+    def test_full_instructional_phrase(self) -> None:
+        """完整就绪指令短语应判定为就绪。"""
+        text = (
+            "1. Gather data\n2. Analyze\n\n"
+            "READY: I am ready to execute the task."
+        )
+        assert response_indicates_ready(text) is True
+
+    def test_not_ready_overrides_full_instructional_phrase(self) -> None:
+        """完整就绪短语后的 NOT READY 应作为最终判定。"""
+        text = "READY: I am ready to execute the task.\nNOT READY"
+        assert response_indicates_ready(text) is False
+
+    def test_bare_ready_on_its_own_line(self) -> None:
+        """独占一行的 READY 应判定为就绪。"""
+        text = "Step 1: brainstorm ideas\nStep 2: refine them\n\nREADY\n"
+        assert response_indicates_ready(text) is True
+
+    def test_bare_ready_with_period(self) -> None:
+        """带句号的独立 READY 标记应判定为就绪。"""
+        assert response_indicates_ready("Plan looks solid.\n\nREADY.") is True
+
+    def test_ready_with_colon_short_form(self) -> None:
+        """带冒号的 READY 简写应判定为就绪。"""
+        assert response_indicates_ready("Details...\nREADY: proceed") is True
+
+    def test_not_ready_is_false(self) -> None:
+        """独立的 NOT READY 标记应判定为未就绪。"""
+        text = "Still missing context.\n\nNOT READY"
+        assert response_indicates_ready(text) is False
+
+    def test_not_ready_does_not_match_as_ready(self) -> None:
+        """NOT READY 中的 READY 子串不应误判为就绪。"""
+        assert response_indicates_ready("NOT READY") is False
+        assert response_indicates_ready("NOT READY.") is False
+
+    def test_mid_sentence_ready_is_not_a_marker(self) -> None:
+        """句子中间的 ready 单词不应视为就绪标记。"""
+        assert (
+            response_indicates_ready("I am ready to begin researching sources.")
+            is False
+        )
+
+    def test_last_marker_wins_after_refinement(self) -> None:
+        """计划细化后应以最后一个 READY 标记为准。"""
+        text = (
+            "Initial thoughts...\nNOT READY\n"
+            "Refined plan with missing pieces filled in.\nREADY"
+        )
+        assert response_indicates_ready(text) is True
+
+    def test_last_marker_can_revert_to_not_ready(self) -> None:
+        """最后一个 NOT READY 标记应撤销先前的就绪判定。"""
+        text = "READY\nActually wait, still incomplete.\nNOT READY"
+        assert response_indicates_ready(text) is False
+
+    def test_empty_and_none_like(self) -> None:
+        """空响应和仅含空白的响应应判定为未就绪。"""
+        assert response_indicates_ready("") is False
+        assert response_indicates_ready("   \n  ") is False
+
+    def test_plan_without_marker_is_not_ready(self) -> None:
+        """不含就绪标记的计划应判定为未就绪。"""
+        assert response_indicates_ready("Just a plan with no conclusion.") is False
+
+    def test_case_insensitive_ready_line(self) -> None:
+        """就绪标记匹配应忽略字母大小写。"""
+        assert response_indicates_ready("plan body\nready") is True
+        assert response_indicates_ready("plan body\nnot ready") is False
+
+
+class TestParsePlanningResponseReady:
+    """Ensure _parse_planning_response uses the shared detector."""
+
+    @pytest.mark.parametrize(
+        ("response", "expected_ready"),
+        [
+            ("Plan...\nREADY: I am ready to execute the task.", True),
+            ("Plan...\nREADY", True),
+            ("Plan...\nNOT READY", False),
+            ("", False),
+        ],
+    )
+    def test_parse_planning_response(self, response: str, expected_ready: bool) -> None:
+        """规划响应解析应复用共享的就绪检测逻辑。"""
+        plan, ready = AgentReasoning._parse_planning_response(response)
+        assert ready is expected_ready
+        if response:
+            assert plan == response
+        else:
+            assert plan == "No plan was generated."