markdown-flow 0.2.30__py3-none-any.whl → 0.2.31__py3-none-any.whl

markdown_flow/__init__.py

@@ -82,5 +82,5 @@ __all__ = [
     "replace_variables_in_text",
 ]
 
-__version__ = "0.2.30"
-# __version__ = "0.2.29-alpha-1"
+__version__ = "0.2.31"
+# __version__ = "0.2.30-alpha-1"

markdown_flow/constants.py

@@ -26,12 +26,6 @@ COMPILED_LAYER3_BUTTON_VALUE_REGEX = re.compile(r"^(.+)//(.+)$")  # Layer 3: Par
 COMPILED_BRACE_VARIABLE_REGEX = re.compile(
     r"(?<!%)\{\{([^}]+)\}\}"  # Match {{variable}} format for replaceable variables
 )
-COMPILED_INTERACTION_CONTENT_RECONSTRUCT_REGEX = re.compile(
-    r"(\?\[[^]]*\.\.\.)([^]]*\])"  # Reconstruct interaction content: prefix + question + suffix
-)
-COMPILED_BRACKETS_CLEANUP_REGEX = re.compile(r"[\[\]()]")
-COMPILED_VARIABLE_REFERENCE_CLEANUP_REGEX = re.compile(r"%\{\{[^}]*\}\}")
-COMPILED_WHITESPACE_CLEANUP_REGEX = re.compile(r"\s+")
 COMPILED_SINGLE_PIPE_SPLIT_REGEX = re.compile(r"(?<!\|)\|(?!\|)")  # Split on single | but not ||
 
 # Document parsing constants (using shared INTERACTION_PATTERN defined above)
@@ -152,28 +146,6 @@ DEFAULT_INTERACTION_PROMPT = """<interaction_translation_rules>
 # Interaction error prompt templates
 DEFAULT_INTERACTION_ERROR_PROMPT = "请将以下错误信息改写得更加友好和个性化，帮助用户理解问题并给出建设性的引导："
 
-# Detailed interaction rendering instructions
-INTERACTION_RENDER_INSTRUCTIONS = """
-核心要求：
-1. **绝对禁止改变问题的含义和方向** - 这是最重要的原则
-2. 只能改变表达方式，不能改变问题的核心内容
-3. 必须保持问题的主体和客体关系不变
-4. 只返回改写后的问题文本，不要包含任何其他内容
-5. 保持专业友好的语气，禁止可爱化表达
-
-关键示例说明：
-✅ 正确改写（保持含义）：
-- "希望我怎么称呼你？" → "请问我应该如何称呼您？"
-- "请输入您的姓名" → "请告诉我您的姓名"
-- "你的年龄是多少？" → "请问您今年多大了？"
-
-❌ 严重错误（改变含义）：
-- "希望我怎么称呼你？" → "你想叫我什么名字？" （方向颠倒）
-- "请输入您的姓名" → "我叫什么好呢？" （主客体颠倒）
-- "你喜欢什么？" → "我应该喜欢什么？" （完全改变意思）
-
-请严格按照以上要求改写，确保不改变问题的原始含义："""
-
 # Interaction error rendering instructions
 INTERACTION_ERROR_RENDER_INSTRUCTIONS = """
 请只返回友好的错误提示，不要包含其他格式或说明。"""
@@ -284,25 +256,63 @@ OUTPUT_INSTRUCTION_EXPLANATION = f"""<preserve_or_translate_instruction>
 """
 
 # Validation task template (merged with system message)
-VALIDATION_TASK_TEMPLATE = """你是一个验证用户输入的助手，请严格按照给定的指令进行验证。
-
-# 任务
-从用户回答中提取相关信息，返回JSON格式结果：
-- 合法：{{"result": "ok", "parse_vars": {{"{target_variable}": "提取的内容"}}}}
-- 不合法：{{"result": "illegal", "reason": "原因"}}
-
-# 输出语言
-- 如果在 <document_context> 中明确要求使用特定语言，则错误信息和原因说明应使用该语言
-- 否则，使用用户输入或问题描述的主要语言"""
-
-# Validation requirements template (lenient general version)
-VALIDATION_REQUIREMENTS_TEMPLATE = """# 提取要求
-1. 仔细阅读上述相关问题，理解这个问题想要获取什么信息
-2. 从用户回答中提取与该问题相关的信息
-3. 如果提供了预定义选项，用户选择这些选项时都应该接受；自定义输入只要是对问题的合理回答即可接受
-4. 对于昵称、姓名、标签等自由文本输入，任何非空的合理表达都应该接受（包括数字、字母、符号、emoji等创意性表达）
-5. 只有当用户回答完全无关、包含不当内容或明显违背常识时才标记为不合法
-6. 宽松验证原则：理解用户意图，接受多样化的合理表达形式"""
+VALIDATION_TASK_TEMPLATE = """你是字符串验证程序，不是对话助手。
+
+你的唯一任务：按后续规则检查输入，输出 JSON：
+{{"result": "ok", "parse_vars": {{"{target_variable}": "用户输入"}}}} 或 {{"result": "illegal", "reason": "原因"}}
+
+严禁输出任何自然语言解释。
+
+# reason 语言
+从 <document_context> 中仅提取语言要求（如"使用英文"、"use English"）
+- 如果有明确语言要求 → reason 使用该语言
+- 否则 → reason 使用用户输入或问题的语言
+"""
+
+# Validation requirements template (极致宽松版本)
+VALIDATION_REQUIREMENTS_TEMPLATE = """# 验证算法（按顺序执行）
+
+步骤 1：空值检查（字符串长度检查）
+
+检查规则：input.trim().length == 0 ?
+- YES → 空
+- NO  → 非空
+
+⚠️ 只要去除首尾空格后字符数 > 0，就是非空
+⚠️ 不判断语义！所有可见字符（a、1、@、中）都计入长度
+⚠️ 示例：
+  - ""      → 长度0 → 空
+  - "  "    → 长度0 → 空
+  - "aa"    → 长度2 → 非空
+  - "@_@"   → 长度3 → 非空
+  - "棒棒糖" → 长度3 → 非空
+
+步骤 2：模糊回答检查
+
+拒绝以下模糊回答："不知道"、"不清楚"、"没有"、"不告诉你"
+
+步骤 3：宗教政治检查
+
+只拒绝明确的宗教政治立场表达（宗教教义、政治口号等）
+地名,地区等（北京、上海等）、普通词汇都不算
+
+步骤 4：输出结果（reason 语言跟随 <document_context> 中的语言要求）
+
+伪代码逻辑：
+  if 空:
+      输出 {{"result": "illegal", "reason": "输入为空（或对应语言的翻译）"}}
+  else if 模糊回答:
+      输出 {{"result": "illegal", "reason": "请提供具体内容（或对应语言的翻译）"}}
+  else if 宗教政治:
+      输出 {{"result": "illegal", "reason": "包含敏感内容（或对应语言的翻译）"}}
+  else:
+      输出 {{"result": "ok", "parse_vars": {{"{target_variable}": "用户输入"}}}}
+
+⚠️ 极致重要：
+- len(去除空格后的输入) > 0 → 必须视为非空
+- 符号、数字、品牌名、地名等都不是"空"，也不是"无效"
+- 默认通过，只在明确违规时才拒绝
+"""
 
 # ========== Error Message Constants ==========
 

markdown_flow/core.py

@@ -97,6 +97,7 @@ class MarkdownFlow:
         self._blocks = None
         self._model: str | None = None
         self._temperature: float | None = None
+        self._enable_text_validation: bool = False  # Default: validation disabled for performance
 
         # Preprocess document: extract code blocks and replace with placeholders
         # This is done once during initialization, similar to Go implementation
@@ -194,6 +195,37 @@ class MarkdownFlow:
         """
         return self._temperature
 
+    def set_text_validation_enabled(self, enabled: bool) -> "MarkdownFlow":
+        """
+        Set whether to enable text input LLM validation.
+
+        Default is False (disabled) for performance and cost optimization.
+        When disabled, text inputs are accepted directly without LLM validation.
+        When enabled, uses ValidationTaskTemplate for LLM validation.
+
+        Affects interaction types:
+        - TEXT_ONLY: Pure text input
+        - BUTTONS_WITH_TEXT: Buttons + text fallback
+        - BUTTONS_MULTI_WITH_TEXT: Multi-select buttons + text fallback
+
+        Args:
+            enabled: True to enable validation, False to disable
+
+        Returns:
+            Self for method chaining
+        """
+        self._enable_text_validation = enabled
+        return self
+
+    def is_text_validation_enabled(self) -> bool:
+        """
+        Check if text input validation is enabled.
+
+        Returns:
+            True if validation is enabled, False otherwise
+        """
+        return self._enable_text_validation
+
     def set_prompt(self, prompt_type: str, value: str | None) -> None:
         """
         Set prompt template.
@@ -558,9 +590,30 @@ class MarkdownFlow:
             # Step 1: Match button values
             matched_values, unmatched_values = self._match_button_values(buttons, target_values)
 
-            # Step 2: If there are unmatched values (custom text), validate with LLM
+            # Step 2: If there are unmatched values (custom text)
             if unmatched_values:
-                # Create user_input for LLM validation (only custom text)
+                # Check if text validation is enabled
+                if not self._enable_text_validation:
+                    # Validation disabled: directly accept unmatched custom text
+                    all_values = matched_values + unmatched_values
+                    result = LLMResult(
+                        content="",
+                        variables={target_variable: all_values},
+                        metadata={
+                            "interaction_type": str(interaction_type),
+                            "matched_button_values": matched_values,
+                            "custom_text_values": unmatched_values,
+                            "validation_bypassed": True,  # Mark validation was skipped
+                        },
+                    )
+                    # Return generator for STREAM mode, direct result for COMPLETE mode
+                    if mode == ProcessMode.STREAM:
+                        def stream_generator():
+                            yield result
+                        return stream_generator()
+                    return result
+
+                # Validation enabled: create user_input for LLM validation (only custom text)
                 custom_input = {target_variable: unmatched_values}
 
                 validation_result = self._process_llm_validation(
@@ -614,7 +667,7 @@ class MarkdownFlow:
                     return stream_merge_generator()
             else:
                 # All values matched buttons, return directly
-                return LLMResult(
+                result = LLMResult(
                     content="",
                     variables={target_variable: matched_values},
                     metadata={
@@ -622,6 +675,12 @@ class MarkdownFlow:
                         "all_matched_buttons": True,
                     },
                 )
+                # Return generator for STREAM mode, direct result for COMPLETE mode
+                if mode == ProcessMode.STREAM:
+                    def stream_generator():
+                        yield result
+                    return stream_generator()
+                return result
 
         if interaction_type in [
             InteractionType.BUTTONS_ONLY,
@@ -640,7 +699,7 @@ class MarkdownFlow:
         if interaction_type == InteractionType.NON_ASSIGNMENT_BUTTON:
             # Non-assignment buttons: ?[Continue] or ?[Continue|Cancel]
             # These buttons don't assign variables, any input completes the interaction
-            return LLMResult(
+            result = LLMResult(
                 content="",  # Empty content indicates interaction complete
                 variables={},  # Non-assignment buttons don't set variables
                 metadata={
@@ -648,10 +707,34 @@ class MarkdownFlow:
                     "user_input": user_input,
                 },
             )
+            # Return generator for STREAM mode, direct result for COMPLETE mode
+            if mode == ProcessMode.STREAM:
+                def stream_generator():
+                    yield result
+                return stream_generator()
+            return result
 
         # Text-only input type: ?[%{{sys_user_nickname}}...question]
-        # Use LLM validation to check if input is relevant to the question
         if target_values:
+            # Check if text validation is enabled
+            if not self._enable_text_validation:
+                # Validation disabled: directly accept all text input
+                result = LLMResult(
+                    content="",
+                    variables={target_variable: target_values},
+                    metadata={
+                        "interaction_type": "text_only",
+                        "validation_bypassed": True,  # Mark validation was skipped
+                    },
+                )
+                # Return generator for STREAM mode, direct result for COMPLETE mode
+                if mode == ProcessMode.STREAM:
+                    def stream_generator():
+                        yield result
+                    return stream_generator()
+                return result
+
+            # Validation enabled: use LLM validation to check if input is relevant to the question
             return self._process_llm_validation(
                 block_index=block_index,
                 user_input=user_input,
@@ -728,7 +811,7 @@ class MarkdownFlow:
         if not target_values:
             if allow_text_input:
                 # Allow empty input for buttons+text mode
-                return LLMResult(
+                result = LLMResult(
                     content="",
                     variables={target_variable: []},
                     metadata={
@@ -736,6 +819,12 @@ class MarkdownFlow:
                         "empty_input": True,
                     },
                 )
+                # Return generator for STREAM mode, direct result for COMPLETE mode
+                if mode == ProcessMode.STREAM:
+                    def stream_generator():
+                        yield result
+                    return stream_generator()
+                return result
             # Pure button mode requires input
             button_displays = [btn["display"] for btn in buttons]
             error_msg = f"Please select from: {', '.join(button_displays)}"
@@ -767,7 +856,7 @@ class MarkdownFlow:
             return self._render_error(error_msg, mode, context)
 
         # Success: return validated values
-        return LLMResult(
+        result = LLMResult(
             content="",
             variables={target_variable: valid_values},
             metadata={
@@ -778,6 +867,12 @@ class MarkdownFlow:
                 "total_input_count": len(target_values),
             },
         )
+        # Return generator for STREAM mode, direct result for COMPLETE mode
+        if mode == ProcessMode.STREAM:
+            def stream_generator():
+                yield result
+            return stream_generator()
+        return result
 
     def _process_llm_validation(
         self,
@@ -991,7 +1086,6 @@ class MarkdownFlow:
             translatable["question"] = interaction_info["question"]
 
         # 转换为 JSON
-        import json
 
         json_str = json.dumps(translatable, ensure_ascii=False)
 
@@ -1041,7 +1135,6 @@ class MarkdownFlow:
         Returns:
             str: 重构后的交互内容
         """
-        import json
 
         # 解析原始 JSON
         try:

markdown_flow/parser/validation.py

@@ -8,10 +8,6 @@ import json
 from typing import Any
 
 from ..constants import (
-    CONTEXT_BUTTON_OPTIONS_TEMPLATE,
-    CONTEXT_CONVERSATION_TEMPLATE,
-    CONTEXT_QUESTION_MARKER,
-    CONTEXT_QUESTION_TEMPLATE,
     VALIDATION_ILLEGAL_DEFAULT_REASON,
     VALIDATION_RESPONSE_ILLEGAL,
     VALIDATION_RESPONSE_OK,

{markdown_flow-0.2.30.dist-info → markdown_flow-0.2.31.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: markdown-flow
-Version: 0.2.30
+Version: 0.2.31
 Summary: An agent library designed to parse and process MarkdownFlow documents
 Project-URL: Homepage, https://github.com/ai-shifu/markdown-flow-agent-py
 Project-URL: Bug Tracker, https://github.com/ai-shifu/markdown-flow-agent-py/issues

{markdown_flow-0.2.30.dist-info → markdown_flow-0.2.31.dist-info}/RECORD

@@ -1,6 +1,6 @@
-markdown_flow/__init__.py,sha256=E-P0SvBLJKQSwj2ZijjjXsDFl9axsrX8oTfTl7YBO7w,2808
-markdown_flow/constants.py,sha256=aroEBhrOGY6JxofRxPFe87vesEJFY1Srm0_jRHVdtig,14274
-markdown_flow/core.py,sha256=bwUdblJPPWqV_Utwn3ijmjpauI_FXcZUlVG30BIa7nw,47831
+markdown_flow/__init__.py,sha256=x92vtEJ3BZbfjRvIIGsV68mLxubg_QqXvqKiK9YQVE0,2808
+markdown_flow/constants.py,sha256=l6I16NLgnu_7AxG8hRcSRBLedQPROvA-qqpWLjmGF8Q,13801
+markdown_flow/core.py,sha256=tpUB8fHe16WCsCtq1WHzMVBsLnvBlbKsBRz1reYH8T8,51927
 markdown_flow/enums.py,sha256=Wr41zt0Ce5b3fyLtOTE2erEVp1n92g9OVaBF_BZg_l8,820
 markdown_flow/exceptions.py,sha256=9sUZ-Jd3CPLdSRqG8Pw7eMm7cv_S3VZM6jmjUU8OhIc,976
 markdown_flow/llm.py,sha256=MJRbXKj35AjLCAhWpFhS07s-m3YU2qO1HOFff05HG2I,2239
@@ -12,13 +12,13 @@ markdown_flow/parser/interaction.py,sha256=T4W7iO-iyNJnpM7SmvOH_DRlLuWSDcFyIrN2f
 markdown_flow/parser/json_parser.py,sha256=78GhyyOjlg0l4UmKKNc4zrg-4pSHzrJEt7VKqbz3uyE,1305
 markdown_flow/parser/output.py,sha256=LgxvH6-RINM50p58miQtw_fHER1JEWDGucHk5-sZ-gk,8087
 markdown_flow/parser/preprocessor.py,sha256=YO2znQo7biYAxZZIO5oyrH4h88LZPIe3SidX7ZEOS88,4877
-markdown_flow/parser/validation.py,sha256=fc5-zL4_vsgFQuQ0BHXlJRH5Vkx102SKJy-H72tpLK8,3647
+markdown_flow/parser/validation.py,sha256=qm_YJ0NNWdFrZZfXL800OWTHGtmkX_b3PVF1te7XmTQ,3515
 markdown_flow/parser/variable.py,sha256=eJLbVOyZT8uYM5eJNv5kHLqdRoNz5iNlxHhhi2oDW94,2986
 markdown_flow/providers/__init__.py,sha256=QMr8H9gxoLr6pWXoAb11oZX_She6KWPxnRips537nQ4,319
 markdown_flow/providers/config.py,sha256=Y4Nihqj3KxI6_RyvVKF_mv4mBoPNXeLgYQgv0FqxQfU,2057
 markdown_flow/providers/openai.py,sha256=KgExRJ8QsCeU_c-Yx3IhxG2hBbYN5uZ-uf0VTMvD1LE,12326
-markdown_flow-0.2.30.dist-info/licenses/LICENSE,sha256=qz3BziejhHPd1xa5eVtYEU5Qp6L2pn4otuj194uGxmc,1069
-markdown_flow-0.2.30.dist-info/METADATA,sha256=LEehzrEIw6Q_TyvGFuAG-m0fPEKgO-QcaLgD8LcvyYM,20686
-markdown_flow-0.2.30.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-markdown_flow-0.2.30.dist-info/top_level.txt,sha256=DpigGvQuIt2L0TTLnDU5sylhiTGiZS7MmAMa2hi-AJs,14
-markdown_flow-0.2.30.dist-info/RECORD,,
+markdown_flow-0.2.31.dist-info/licenses/LICENSE,sha256=qz3BziejhHPd1xa5eVtYEU5Qp6L2pn4otuj194uGxmc,1069
+markdown_flow-0.2.31.dist-info/METADATA,sha256=isqtJHsTgU_2KPpmmPfnNnLgzBCeP4PsNTdGyUuNf8I,20686
+markdown_flow-0.2.31.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+markdown_flow-0.2.31.dist-info/top_level.txt,sha256=DpigGvQuIt2L0TTLnDU5sylhiTGiZS7MmAMa2hi-AJs,14
+markdown_flow-0.2.31.dist-info/RECORD,,