diff --git a/ai_test_generator/core/llm_client.py b/ai_test_generator/core/llm_client.py
index a74f090..d2030a2 100644
--- a/ai_test_generator/core/llm_client.py
+++ b/ai_test_generator/core/llm_client.py
@@ -1,58 +1,204 @@
+"""
+LLM 客户端
+══════════════════════════════════════════════════════════════
+变更：
+  - generate_test_cases() 新增 param_constraint 参数
+  - 分批调用时将约束透传给 PromptBuilder.build_batch_prompt()
+"""
+
+from __future__ import annotations
 import json
-import re
 import logging
+import re
+import time
+from typing import Any
+
 from openai import OpenAI
 from config import config
 
 logger = logging.getLogger(__name__)
 
 
-class LLMClient:
-    """封装 LLM API 调用（OpenAI 兼容接口）"""
+# ══════════════════════════════════════════════════════════════
+# JSON 健壮解析工具
+# ══════════════════════════════════════════════════════════════
 
-    def __init__(self):
-        self.client = OpenAI(
-            api_key=config.LLM_API_KEY,
-            base_url=config.LLM_BASE_URL,
-        )
+class RobustJSONParser:
 
-    def generate_test_cases(self, system_prompt: str, user_prompt: str) -> list[dict]:
-        logger.info(f"Calling LLM: model={config.LLM_MODEL}")
-        logger.debug(f"--- USER PROMPT ---\n{user_prompt}\n---")
+    def parse(self, text: str) -> list[dict]:
+        text = text.strip()
+        text = re.sub(r'^```(?:json)?\s*', '', text, flags=re.MULTILINE)
+        text = re.sub(r'\s*```$',          '', text, flags=re.MULTILINE)
+        text = text.strip()
 
-        response = self.client.chat.completions.create(
-            model=config.LLM_MODEL,
-            temperature=config.LLM_TEMPERATURE,
-            max_tokens=config.LLM_MAX_TOKENS,
-            messages=[
-                {"role": "system", "content": system_prompt},
-                {"role": "user",   "content": user_prompt},
-            ],
-        )
-        raw = response.choices[0].message.content
-        logger.debug(f"--- LLM RAW RESPONSE ---\n{raw[:800]}\n---")
-        return self._parse_json(raw)
-
-    # ── 解析 ──────────────────────────────────────────────────
-
-    def _parse_json(self, content: str) -> list[dict]:
-        content = content.strip()
-        # 去除可能的 markdown 代码块
-        content = re.sub(r"^```(?:json)?\s*", "", content, flags=re.MULTILINE)
-        content = re.sub(r"\s*```\s*$", "", content, flags=re.MULTILINE)
-        content = content.strip()
+        start = text.find('[')
+        if start == -1:
+            logger.warning("LLM 响应中未找到 JSON 数组。")
+            return []
+        text = text[start:]
 
         try:
-            data = json.loads(content)
+            result = json.loads(text)
+            if isinstance(result, list):
+                return result
         except json.JSONDecodeError:
-            # 尝试提取第一个 JSON 数组
-            match = re.search(r"\[.*\]", content, re.DOTALL)
-            if match:
-                data = json.loads(match.group())
-            else:
-                logger.error(f"Cannot parse LLM response as JSON:\n{content[:600]}")
-                raise
+            pass
 
-        if not isinstance(data, list):
-            raise ValueError(f"Expected JSON array, got {type(data)}")
-        return data
\ No newline at end of file
+        text = self._fix_truncated(text)
+        try:
+            result = json.loads(text)
+            if isinstance(result, list):
+                logger.warning("JSON 截断修复成功，部分用例可能丢失。")
+                return result
+        except json.JSONDecodeError as e:
+            logger.error(f"JSON 解析失败（修复后仍无效）：{e}")
+            return []
+
+    def _fix_truncated(self, text: str) -> str:
+        last = text.rfind('},')
+        if last == -1:
+            last = text.rfind('}')
+        if last == -1:
+            return text
+        return text[:last + 1] + ']'
+
+
+# ══════════════════════════════════════════════════════════════
+# LLM 客户端
+# ══════════════════════════════════════════════════════════════
+
+class LLMClient:
+
+    DEFAULT_BATCH_SIZE = 10
+
+    def __init__(self):
+        self.client         = OpenAI(
+            api_key=config.LLM_API_KEY,
+            base_url=getattr(config, "LLM_BASE_URL", None),
+        )
+        self.model          = config.LLM_MODEL
+        self.max_tokens     = getattr(config, "LLM_MAX_TOKENS",       8192)
+        self.batch_size     = getattr(config, "LLM_BATCH_SIZE",       self.DEFAULT_BATCH_SIZE)
+        self.retry          = getattr(config, "LLM_RETRY",            3)
+        self.retry_delay    = getattr(config, "LLM_RETRY_DELAY",      5)
+        self.batch_interval = getattr(config, "LLM_BATCH_INTERVAL",   1)
+        self._parser        = RobustJSONParser()
+
+        from core.prompt_builder import PromptBuilder
+        self._prompt_builder = PromptBuilder()
+
+    # ── 主入口 ────────────────────────────────────────────────
+
+    def generate_test_cases(
+        self,
+        system_prompt:    str,
+        user_prompt:      str,
+        iface_summaries:  list[dict] | None = None,
+        requirements:     list[str]  | None = None,
+        project_header:   str = "",
+        param_constraint: "ParamConstraint | None" = None,   # ← 新增
+    ) -> list[dict]:
+        """
+        生成测试用例。
+
+        方式 A（单次）：传 user_prompt。
+        方式 B（分批）：传 iface_summaries + requirements，推荐大规模场景。
+
+        param_constraint：由 RequirementParser 解析出的全局参数约束，
+                          有值时自动注入参数数据集到每批 prompt。
+        """
+        if iface_summaries is not None and requirements is not None:
+            return self._generate_batched(
+                system_prompt, iface_summaries, requirements,
+                project_header, param_constraint,
+            )
+        return self._call_with_retry(system_prompt, user_prompt)
+
+    # ── 分批调用 ──────────────────────────────────────────────
+
+    def _generate_batched(
+        self,
+        system_prompt:    str,
+        iface_summaries:  list[dict],
+        requirements:     list[str],
+        project_header:   str,
+        param_constraint: Any,
+    ) -> list[dict]:
+        total   = len(iface_summaries)
+        batches = self._make_batches(iface_summaries, self.batch_size)
+        all_cases: list[dict] = []
+
+        logger.info(
+            f"分批模式：共 {total} 个接口 → "
+            f"{len(batches)} 批 × 每批最多 {self.batch_size} 个"
+        )
+        if param_constraint and param_constraint.has_param_directive:
+            logger.info(
+                f"参数约束已启用：{param_constraint}"
+            )
+
+        for idx, batch in enumerate(batches, 1):
+            names = [i.get("name", "?") for i in batch]
+            logger.info(f"  第 {idx}/{len(batches)} 批：{names}")
+
+            user_prompt = self._prompt_builder.build_batch_prompt(
+                batch=batch,
+                requirements=requirements,
+                project_header=project_header,
+                param_constraint=param_constraint,   # ← 透传
+            )
+
+            cases = self._call_with_retry(system_prompt, user_prompt)
+            logger.info(f"  第 {idx} 批 → 生成 {len(cases)} 个测试用例")
+            all_cases.extend(cases)
+
+            if idx < len(batches):
+                time.sleep(self.batch_interval)
+
+        logger.info(f"全部批次完成，共生成 {len(all_cases)} 个测试用例")
+        return all_cases
+
+    # ── 单次调用（含重试）────────────────────────────────────
+
+    def _call_with_retry(
+        self,
+        system_prompt: str,
+        user_prompt:   str,
+    ) -> list[dict]:
+        last_error: Exception | None = None
+
+        for attempt in range(1, self.retry + 1):
+            try:
+                logger.debug(f"LLM 调用第 {attempt}/{self.retry} 次 …")
+                response = self.client.chat.completions.create(
+                    model=self.model,
+                    max_tokens=self.max_tokens,
+                    messages=[
+                        {"role": "system", "content": system_prompt},
+                        {"role": "user",   "content": user_prompt},
+                    ],
+                )
+                raw = response.choices[0].message.content or ""
+                logger.debug(f"LLM 响应长度：{len(raw)} 字符")
+
+                cases = self._parser.parse(raw)
+                if cases:
+                    return cases
+
+                logger.warning(f"第 {attempt} 次调用返回空结果，准备重试 …")
+
+            except Exception as e:
+                last_error = e
+                logger.warning(f"第 {attempt} 次调用失败：{e}")
+
+            if attempt < self.retry:
+                time.sleep(self.retry_delay)
+
+        logger.error(
+            f"已重试 {self.retry} 次，全部失败。最后一次错误：{last_error}"
+        )
+        return []
+
+    @staticmethod
+    def _make_batches(items: list[Any], size: int) -> list[list[Any]]:
+        return [items[i:i + size] for i in range(0, len(items), size)]
\ No newline at end of file
diff --git a/ai_test_generator/core/param_generator.py b/ai_test_generator/core/param_generator.py
new file mode 100644
index 0000000..da61601
--- /dev/null
+++ b/ai_test_generator/core/param_generator.py
@@ -0,0 +1,393 @@
+"""
+测试参数生成器
+══════════════════════════════════════════════════════════════
+职责：
+  根据接口参数的数据类型和参数生成策略，
+  自动生成覆盖正常值、边界值、异常值的测试数据集。
+
+支持的数据类型：
+  string / str
+  integer / int / number
+  float / double / decimal
+  boolean / bool
+  array / list
+  object / dict / map
+
+输出格式（供注入 prompt）：
+  {
+    "param_name": {
+      "type": "string",
+      "groups": [
+        {"label": "正常值",   "value": "hello",  "category": "normal"},
+        {"label": "空字符串", "value": "",        "category": "boundary"},
+        {"label": "超长字符串","value": "a"*256,  "category": "exception"},
+        ...
+      ]
+    }
+  }
+"""
+
+from __future__ import annotations
+from dataclasses import dataclass, field
+from typing import Any
+
+from core.requirement_parser import ParamStrategy
+
+
+# ══════════════════════════════════════════════════════════════
+# 数据结构
+# ══════════════════════════════════════════════════════════════
+
+@dataclass
+class ParamGroup:
+    """单组测试参数。"""
+    label:    str        # 人类可读的描述，如"空字符串"
+    value:    Any        # 实际参数值
+    category: str        # "normal" | "boundary" | "exception" | "equivalence"
+
+    def to_dict(self) -> dict:
+        return {
+            "label":    self.label,
+            "value":    self.value,
+            "category": self.category,
+        }
+
+
+@dataclass
+class ParamDataSet:
+    """单个参数的完整测试数据集。"""
+    param_name: str
+    param_type: str
+    groups:     list[ParamGroup] = field(default_factory=list)
+
+    def to_dict(self) -> dict:
+        return {
+            "type":   self.param_type,
+            "groups": [g.to_dict() for g in self.groups],
+        }
+
+    @property
+    def normal_values(self) -> list[Any]:
+        return [g.value for g in self.groups if g.category == "normal"]
+
+    @property
+    def boundary_values(self) -> list[Any]:
+        return [g.value for g in self.groups if g.category == "boundary"]
+
+    @property
+    def exception_values(self) -> list[Any]:
+        return [g.value for g in self.groups if g.category == "exception"]
+
+
+# ══════════════════════════════════════════════════════════════
+# 各类型测试数据库
+# ══════════════════════════════════════════════════════════════
+
+# ── string ────────────────────────────────────────────────────
+_STRING_NORMAL: list[ParamGroup] = [
+    ParamGroup("普通字符串",     "hello",                "normal"),
+    ParamGroup("中文字符串",     "测试数据",              "normal"),
+    ParamGroup("字母数字混合",   "user123",              "normal"),
+]
+_STRING_BOUNDARY: list[ParamGroup] = [
+    ParamGroup("空字符串",       "",                     "boundary"),
+    ParamGroup("单字符",         "a",                    "boundary"),
+    ParamGroup("最大长度(255)",  "a" * 255,              "boundary"),
+    ParamGroup("恰好256字符",    "a" * 256,              "boundary"),
+    ParamGroup("含空格",         "hello world",          "boundary"),
+    ParamGroup("首尾空格",       "  hello  ",            "boundary"),
+]
+_STRING_EXCEPTION: list[ParamGroup] = [
+    ParamGroup("None/null",      None,                   "exception"),
+    ParamGroup("超长字符串",     "a" * 1000,             "exception"),
+    ParamGroup("纯空格",         "   ",                  "exception"),
+    ParamGroup("特殊字符",       "!@#$%^&*()",           "exception"),
+    ParamGroup("换行符",         "line1\nline2",         "exception"),
+    ParamGroup("SQL注入",        "' OR '1'='1",          "exception"),
+    ParamGroup("XSS注入",        "<script>alert(1)</script>", "exception"),
+    ParamGroup("Unicode特殊字符","你好\u0000世界",        "exception"),
+    ParamGroup("数字类型误传",   12345,                  "exception"),
+]
+
+# ── integer ───────────────────────────────────────────────────
+_INTEGER_NORMAL: list[ParamGroup] = [
+    ParamGroup("典型正整数",     1,                      "normal"),
+    ParamGroup("较大正整数",     100,                    "normal"),
+]
+_INTEGER_BOUNDARY: list[ParamGroup] = [
+    ParamGroup("零",             0,                      "boundary"),
+    ParamGroup("正边界值1",      1,                      "boundary"),
+    ParamGroup("负边界值-1",     -1,                     "boundary"),
+    ParamGroup("最大值(int32)",  2_147_483_647,          "boundary"),
+    ParamGroup("最小值(int32)",  -2_147_483_648,         "boundary"),
+    ParamGroup("最大值+1",       2_147_483_648,          "boundary"),
+]
+_INTEGER_EXCEPTION: list[ParamGroup] = [
+    ParamGroup("None/null",      None,                   "exception"),
+    ParamGroup("字符串类型",     "abc",                  "exception"),
+    ParamGroup("浮点数",         3.14,                   "exception"),
+    ParamGroup("布尔值true",     True,                   "exception"),
+    ParamGroup("空字符串",       "",                     "exception"),
+    ParamGroup("超大整数",       10 ** 20,               "exception"),
+]
+
+# ── float ─────────────────────────────────────────────────────
+_FLOAT_NORMAL: list[ParamGroup] = [
+    ParamGroup("典型浮点数",     3.14,                   "normal"),
+    ParamGroup("整数形式",       1.0,                    "normal"),
+]
+_FLOAT_BOUNDARY: list[ParamGroup] = [
+    ParamGroup("零",             0.0,                    "boundary"),
+    ParamGroup("极小正数",       1e-10,                  "boundary"),
+    ParamGroup("极大正数",       1e10,                   "boundary"),
+    ParamGroup("负数",           -3.14,                  "boundary"),
+    ParamGroup("负无穷",         float('-inf'),          "boundary"),
+    ParamGroup("正无穷",         float('inf'),           "boundary"),
+]
+_FLOAT_EXCEPTION: list[ParamGroup] = [
+    ParamGroup("None/null",      None,                   "exception"),
+    ParamGroup("NaN",            float('nan'),           "exception"),
+    ParamGroup("字符串类型",     "abc",                  "exception"),
+    ParamGroup("空字符串",       "",                     "exception"),
+]
+
+# ── boolean ───────────────────────────────────────────────────
+_BOOLEAN_NORMAL: list[ParamGroup] = [
+    ParamGroup("true",           True,                   "normal"),
+    ParamGroup("false",          False,                  "normal"),
+]
+_BOOLEAN_BOUNDARY: list[ParamGroup] = [
+    ParamGroup("整数1",          1,                      "boundary"),
+    ParamGroup("整数0",          0,                      "boundary"),
+]
+_BOOLEAN_EXCEPTION: list[ParamGroup] = [
+    ParamGroup("None/null",      None,                   "exception"),
+    ParamGroup("字符串'true'",   "true",                 "exception"),
+    ParamGroup("字符串'false'",  "false",                "exception"),
+    ParamGroup("字符串'yes'",    "yes",                  "exception"),
+    ParamGroup("随机字符串",     "abc",                  "exception"),
+]
+
+# ── array ─────────────────────────────────────────────────────
+_ARRAY_NORMAL: list[ParamGroup] = [
+    ParamGroup("普通数组",       [1, 2, 3],              "normal"),
+    ParamGroup("字符串数组",     ["a", "b", "c"],        "normal"),
+]
+_ARRAY_BOUNDARY: list[ParamGroup] = [
+    ParamGroup("空数组",         [],                     "boundary"),
+    ParamGroup("单元素数组",     [1],                    "boundary"),
+    ParamGroup("大数组(100元素)", list(range(100)),      "boundary"),
+    ParamGroup("含None元素",     [1, None, 3],           "boundary"),
+]
+_ARRAY_EXCEPTION: list[ParamGroup] = [
+    ParamGroup("None/null",      None,                   "exception"),
+    ParamGroup("字符串类型",     "not_an_array",         "exception"),
+    ParamGroup("整数类型",       123,                    "exception"),
+    ParamGroup("嵌套过深",       [[[[[1]]]]],            "exception"),
+]
+
+# ── object ────────────────────────────────────────────────────
+_OBJECT_NORMAL: list[ParamGroup] = [
+    ParamGroup("完整字段对象",   {"id": 1, "name": "test"}, "normal"),
+]
+_OBJECT_BOUNDARY: list[ParamGroup] = [
+    ParamGroup("空对象",         {},                     "boundary"),
+    ParamGroup("仅必填字段",     {"id": 1},              "boundary"),
+    ParamGroup("含额外字段",     {"id": 1, "extra": "x"}, "boundary"),
+]
+_OBJECT_EXCEPTION: list[ParamGroup] = [
+    ParamGroup("None/null",      None,                   "exception"),
+    ParamGroup("字符串类型",     "not_an_object",        "exception"),
+    ParamGroup("数组类型",       [1, 2, 3],              "exception"),
+    ParamGroup("字段值类型错误", {"id": "not_int"},      "exception"),
+]
+
+# 类型 → 数据映射表
+_TYPE_DATA: dict[str, dict[str, list[ParamGroup]]] = {
+    "string":  {"normal": _STRING_NORMAL,  "boundary": _STRING_BOUNDARY,  "exception": _STRING_EXCEPTION},
+    "integer": {"normal": _INTEGER_NORMAL, "boundary": _INTEGER_BOUNDARY, "exception": _INTEGER_EXCEPTION},
+    "float":   {"normal": _FLOAT_NORMAL,   "boundary": _FLOAT_BOUNDARY,   "exception": _FLOAT_EXCEPTION},
+    "boolean": {"normal": _BOOLEAN_NORMAL, "boundary": _BOOLEAN_BOUNDARY, "exception": _BOOLEAN_EXCEPTION},
+    "array":   {"normal": _ARRAY_NORMAL,   "boundary": _ARRAY_BOUNDARY,   "exception": _ARRAY_EXCEPTION},
+    "object":  {"normal": _OBJECT_NORMAL,  "boundary": _OBJECT_BOUNDARY,  "exception": _OBJECT_EXCEPTION},
+}
+
+# 类型别名归一化
+_TYPE_ALIAS: dict[str, str] = {
+    "str":     "string",  "text":    "string",  "varchar": "string",
+    "int":     "integer", "long":    "integer", "number":  "integer",
+    "float":   "float",   "double":  "float",   "decimal": "float",
+    "bool":    "boolean",
+    "list":    "array",
+    "dict":    "object",  "map":     "object",  "json":    "object",
+}
+
+
+# ══════════════════════════════════════════════════════════════
+# 参数生成器
+# ══════════════════════════════════════════════════════════════
+
+class ParamGenerator:
+    """
+    根据参数类型和策略生成测试数据集。
+    """
+
+    def generate(
+        self,
+        param_name:  str,
+        param_type:  str,
+        strategies:  list[ParamStrategy],
+        min_groups:  int = 2,
+    ) -> ParamDataSet:
+        """
+        生成单个参数的测试数据集。
+
+        Args:
+          param_name : 参数名
+          param_type : 数据类型（支持别名）
+          strategies : 需要覆盖的策略列表
+          min_groups : 最少生成的参数组数
+
+        Returns:
+          ParamDataSet
+        """
+        norm_type = self._normalize_type(param_type)
+        type_data = _TYPE_DATA.get(norm_type, _TYPE_DATA["string"])
+
+        groups: list[ParamGroup] = []
+
+        # 始终包含正常值
+        groups.extend(type_data["normal"])
+
+        # 按策略追加
+        for strategy in strategies:
+            if strategy == ParamStrategy.BOUNDARY:
+                groups.extend(type_data["boundary"])
+            elif strategy == ParamStrategy.EXCEPTION:
+                groups.extend(type_data["exception"])
+            elif strategy == ParamStrategy.EQUIVALENCE:
+                # 等价类：正常值 + 部分边界值
+                groups.extend(type_data["boundary"][:3])
+            elif strategy == ParamStrategy.RANDOM:
+                groups.extend(self._random_groups(norm_type))
+
+        # 去重（按 label）
+        seen_labels: set[str] = set()
+        unique_groups: list[ParamGroup] = []
+        for g in groups:
+            if g.label not in seen_labels:
+                seen_labels.add(g.label)
+                unique_groups.append(g)
+
+        # 补齐到 min_groups
+        if len(unique_groups) < min_groups:
+            extra = type_data["boundary"] + type_data["exception"]
+            for g in extra:
+                if g.label not in seen_labels and len(unique_groups) < min_groups:
+                    seen_labels.add(g.label)
+                    unique_groups.append(g)
+
+        return ParamDataSet(
+            param_name=param_name,
+            param_type=norm_type,
+            groups=unique_groups,
+        )
+
+    def generate_for_interface(
+        self,
+        interface_summary: dict,
+        strategies:        list[ParamStrategy],
+        min_groups:        int = 2,
+    ) -> dict[str, ParamDataSet]:
+        """
+        为一个接口的所有入参生成测试数据集。
+
+        Args:
+          interface_summary : parser.to_summary_dict() 中的单个接口 dict
+          strategies        : 策略列表
+          min_groups        : 最少组数
+
+        Returns:
+          { param_name: ParamDataSet }
+        """
+        result: dict[str, ParamDataSet] = {}
+        params = interface_summary.get("params", {})
+
+        for param_name, param_info in params.items():
+            if isinstance(param_info, dict):
+                param_type = param_info.get("type", "string")
+                inout      = param_info.get("inout", "in")
+            else:
+                param_type = str(param_info)
+                inout      = "in"
+
+            # 只对入参（in / inout）生成测试数据
+            if inout not in ("in", "inout"):
+                continue
+
+            result[param_name] = self.generate(
+                param_name=param_name,
+                param_type=param_type,
+                strategies=strategies,
+                min_groups=min_groups,
+            )
+
+        return result
+
+    # ── 工具 ──────────────────────────────────────────────────
+
+    @staticmethod
+    def _normalize_type(t: str) -> str:
+        t = t.lower().strip()
+        return _TYPE_ALIAS.get(t, t if t in _TYPE_DATA else "string")
+
+    @staticmethod
+    def _random_groups(norm_type: str) -> list[ParamGroup]:
+        """生成少量随机值（补充策略）。"""
+        import random, string
+        if norm_type == "string":
+            val = ''.join(random.choices(string.ascii_letters, k=8))
+            return [ParamGroup(f"随机字符串({val})", val, "equivalence")]
+        if norm_type == "integer":
+            val = random.randint(-1000, 1000)
+            return [ParamGroup(f"随机整数({val})", val, "equivalence")]
+        if norm_type == "float":
+            val = round(random.uniform(-100, 100), 4)
+            return [ParamGroup(f"随机浮点数({val})", val, "equivalence")]
+        return []
+
+    def to_prompt_text(
+        self,
+        datasets: dict[str, ParamDataSet],
+        min_groups: int,
+    ) -> str:
+        """
+        将参数数据集转换为注入 prompt 的中文说明文本。
+        """
+        if not datasets:
+            return ""
+
+        lines = [
+            f"### 参数测试数据集（每个参数至少覆盖 {min_groups} 组）",
+            "",
+        ]
+        for param_name, ds in datasets.items():
+            lines.append(f"**参数：{param_name}**（类型：{ds.param_type}）")
+            for g in ds.groups:
+                category_label = {
+                    "normal":      "正常值",
+                    "boundary":    "边界值",
+                    "exception":   "异常值",
+                    "equivalence": "等价类",
+                }.get(g.category, g.category)
+                lines.append(
+                    f"  - [{category_label}] {g.label}：`{repr(g.value)}`"
+                )
+            lines.append("")
+
+        lines.append(
+            f"> 请从以上数据集中选取参数值组合，"
+            f"确保生成的测试用例总数不少于 **{min_groups} 组**，"
+            f"并覆盖正常值、边界值和异常值场景。"
+        )
+        return "\n".join(lines)
\ No newline at end of file
diff --git a/ai_test_generator/core/prompt_builder.py b/ai_test_generator/core/prompt_builder.py
index 1612df2..2ab07e8 100644
--- a/ai_test_generator/core/prompt_builder.py
+++ b/ai_test_generator/core/prompt_builder.py
@@ -1,167 +1,332 @@
 """
-提示词构建器
-升级点：
-  - 传递项目 description
-  - function 协议：source_file(.py) → module_path → from <module> import <func>
-  - HTTP 协议：full_url = url(base) + name(path)
-  - parameters 统一字段，inout 区分输入/输出
+提示词构建器（中文版）
+══════════════════════════════════════════════════════════════
+变更说明：
+  - 新增 build_param_directive_section()：将参数约束注入 System Prompt
+  - 新增 build_param_dataset_section()：将参数数据集注入 User Prompt
+  - build_batch_prompt / build_user_prompt 支持传入参数数据集
 """
 
 import json
 from core.parser import InterfaceInfo, InterfaceParser
+from core.requirement_parser import ParamConstraint, ParamStrategy
+from core.param_generator import ParamGenerator, ParamDataSet
 
-SYSTEM_PROMPT = """
-You are a senior software test engineer. Generate test cases based on the provided
-interface descriptions and test requirements.
+
+# ══════════════════════════════════════════════════════════════
+# System Prompt（中文）
+# ══════════════════════════════════════════════════════════════
+
+_SYSTEM_PROMPT_BASE = """
+你是一名资深软件测试工程师，擅长根据接口描述和测试需求生成高质量的测试用例。
 
 ═══════════════════════════════════════════════════════
-OUTPUT FORMAT
-Return ONLY a valid JSON array. No markdown fences. No extra text.
+【输出格式要求】
 ═══════════════════════════════════════════════════════
-Each element = ONE test case:
+- 只输出一个合法的 JSON 数组，不要包含任何 markdown 代码块（```）
+- 不要在 JSON 前后添加任何解释性文字
+- 每个数组元素代表一个测试用例，结构如下：
+
 {
-  "test_id"        : "unique id, e.g. TC_001",
-  "test_name"      : "short descriptive name",
-  "description"    : "what this test verifies",
-  "requirement"    : "the original requirement text this case maps to",
-  "requirement_id" : "e.g. REQ.01",
+  "test_id"         : "唯一编号，如 TC_001",
+  "test_name"       : "简短的测试名称",
+  "description"     : "本用例验证的内容",
+  "requirement"     : "对应的原始需求文本",
+  "requirement_id"  : "需求编号，如 REQ.01（从接口描述中的 requirement_id 字段获取）",
   "steps": [
     {
-      "step_no"        : 1,
-      "interface_name" : "function or endpoint name",
-      "protocol"       : "http or function",
-      "url"            : "source_file (function) or full_url (http)",
-      "purpose"        : "why this step is needed",
+      "step_no"         : 1,
+      "interface_name"  : "接口名称或函数名",
+      "protocol"        : "http 或 function",
+      "url"             : "function 协议填 source_file，http 协议填 full_url",
+      "purpose"         : "本步骤的目的",
       "input": {
-        // Only parameters with inout = "in" or "inout"
-        "param_name": <value>
+        "参数名": "参数值"
       },
-      "use_output_of": {           // optional
+      "use_output_of": {
         "step_no"  : 1,
-        "field"    : "user_id",
-        "as_param" : "user_id"
+        "field"    : "上一步返回值中的字段名",
+        "as_param" : "作为本步骤的参数名"
       },
       "assertions": [
         {
-          "field"    : "field_name or 'return' or 'exception'",
-          "operator" : "eq|ne|gt|lt|gte|lte|in|not_null|contains|raised|not_raised",
-          "expected" : <value>,
-          "message"  : "human readable description"
+          "field"    : "断言的字段名，或 'return'（整体返回值）或 'exception'（异常）",
+          "operator" : "eq | ne | gt | lt | gte | lte | in | not_null | contains | raised | not_raised",
+          "expected" : "期望值",
+          "message"  : "断言说明"
         }
       ]
     }
   ],
-  "test_data_notes" : "explanation of auto-generated test data",
-  "test_code"       : "<complete Python script — see rules below>"
+  "test_data_notes" : "测试数据说明（说明本用例使用了哪类参数：正常值/边界值/异常值）",
+  "test_code"       : "完整可运行的 Python 测试脚本（见下方规则）"
 }
 
 ═══════════════════════════════════════════════════════
-TEST CODE RULES
+【测试代码编写规则】
 ═══════════════════════════════════════════════════════
-1. Complete, runnable Python script. No external test framework.
-2. Allowed imports: standard library + `requests` + the actual module under test.
+1. test_code 必须是完整、可直接运行的 Python 脚本
+2. 不得使用 unittest 或 pytest 框架
+3. 只允许导入：Python 标准库、requests、被测模块
 
-── FUNCTION INTERFACES ────────────────────────────────
-3. Each function interface has:
-     "source_file"  : e.g. "create_user.py"
-     "module_path"  : e.g. "create_user"   (derived from source_file)
-4. Import the REAL function using module_path:
+── function 协议接口 ──────────────────────────────────
+4. 使用 module_path 字段导入真实函数，若导入失败则使用桩函数：
      try:
          from <module_path> import <function_name>
      except ImportError:
-         # Stub fallback — simulate on_success for positive tests,
-         # on_failure / raise Exception for negative tests
-         def <function_name>(<in_params>):
-             return <on_success_value>   # or raise Exception(...)
-5. Call the function with ONLY "in"/"inout" parameters.
-6. Capture the return value; assert against on_success or on_failure structure.
-7. If on_failure.value contains "exception" or "raises":
-     - In negative tests: wrap call in try/except, assert exception IS raised.
-     - Use field="exception", operator="raised", expected="Exception".
+         def <function_name>(<入参列表>):
+             return <on_success_value>   # 桩函数，仅供结构验证
+5. 调用函数时只传 inout 为 "in" 或 "inout" 的参数
+6. 若 on_failure 包含 "exception"，负向测试需用 try/except 捕获异常
 
-── HTTP INTERFACES ────────────────────────────────────
-8. Each HTTP interface has:
-     "full_url" : complete URL, e.g. "http://127.0.0.1/api/delete_user"
-     "method"   : "get" | "post" | "put" | "delete" etc.
-9. Send request:
+── http 协议接口 ──────────────────────────────────────
+7. 使用 full_url 字段作为请求地址：
      resp = requests.<method>("<full_url>", json=<input_dict>)
-10. Assert on resp.status_code and resp.json() fields.
+8. 断言 resp.status_code 和 resp.json() 中的字段
 
-── MULTI-STEP ─────────────────────────────────────────
-11. Execute steps in order; extract fields from previous step's return value
-    and pass to next step via use_output_of mapping.
-
-── STRUCTURED OUTPUT (REQUIRED) ───────────────────────
-12. After EACH step, print:
-      ##STEP_RESULT## {"step_no":<n>,"interface_name":"...","status":"PASS|FAIL",
-        "assertions":[{"field":"...","operator":"...","expected":...,
-                       "actual":...,"passed":true|false,"message":"..."}]}
-13. Final line must be:
-      PASS: <summary>   or   FAIL: <summary>
-14. Wrap entire test body in try/except; print FAIL on any unhandled exception.
-15. Do NOT use unittest or pytest.
+── 结构化输出（必须遵守）─────────────────────────────
+9. 每个步骤执行后，必须在标准输出打印如下格式的一行：
+     ##STEP_RESULT## {"step_no":<n>,"interface_name":"...","status":"PASS 或 FAIL",
+       "assertions":[{"field":"...","operator":"...","expected":...,
+                      "actual":...,"passed":true 或 false,"message":"..."}]}
+10. 脚本最后一行必须是：
+      PASS: <摘要>   或   FAIL: <摘要>
+11. 整个脚本主体用 try/except 包裹，捕获未处理异常时打印 FAIL
 
 ═══════════════════════════════════════════════════════
-ASSERTION RULES BY RETURN TYPE
+【断言编写规则】
 ═══════════════════════════════════════════════════════
-return.type = "dict"    → assert each key in on_success.value (positive)
-                          or on_failure.value (negative)
-return.type = "boolean" → field="return", operator="eq", expected=true/false
-return.type = "integer" → field="return", operator="eq"/"gt"/"lt"
-on_failure  = exception → field="exception", operator="raised"
+- 返回值为 dict  → 对 on_success / on_failure 中每个 key 单独断言
+- 返回值为 bool  → field="return"，operator="eq"，expected=true 或 false
+- 返回值为 int   → field="return"，operator="eq" / "gt" / "lt"
+- 预期抛出异常   → field="exception"，operator="raised"
+- 预期不抛出异常 → field="exception"，operator="not_raised"
 
 ═══════════════════════════════════════════════════════
-COVERAGE GUIDELINES
+【覆盖度要求】
 ═══════════════════════════════════════════════════════
-- Per requirement: at least 1 positive + 1 negative test case.
-- Negative test_name/description MUST contain "negative" or "invalid".
-- Multi-interface requirements: single multi-step test case.
-- Cover ALL "in"/"inout" parameters (including optional ones).
-- Assert ALL fields described in on_success (positive) and on_failure (negative).
-- For "out" parameters: verify their values in assertions after the call.
+- 每条需求至少生成 1 个正向用例 + 1 个负向用例
+- 负向用例的 test_name 必须包含"负向"或"无效"
+- 覆盖接口的所有 inout 为 "in" 或 "inout" 的参数
+- 正向用例断言 on_success 的所有字段
+- 负向用例断言 on_failure 的所有字段
+"""
+
+# 参数生成规则段落（有参数约束时追加到 System Prompt）
+_PARAM_DIRECTIVE_TEMPLATE = """
+═══════════════════════════════════════════════════════
+【参数生成规则（本次任务特殊要求）】
+═══════════════════════════════════════════════════════
+{directives}
+- test_data_notes 字段必须说明本用例使用的参数类别（正常值/边界值/异常值）及具体值
+- 每个测试用例的 input 字段中的参数值，必须从下方"参数测试数据集"中选取
+- 禁止在 input 中使用未在数据集中出现的随机值
 """
 
 
+# ══════════════════════════════════════════════════════════════
+# PromptBuilder
+# ══════════════════════════════════════════════════════════════
+
 class PromptBuilder:
 
-    def get_system_prompt(self) -> str:
-        return SYSTEM_PROMPT.strip()
+    def __init__(self):
+        self._param_gen = ParamGenerator()
+
+    # ── System Prompt ────────────────────────────────────────
+
+    def get_system_prompt(
+        self,
+        global_constraint: ParamConstraint = None,
+    ) -> str:
+        """
+        构建 System Prompt。
+        若存在全局参数约束，追加参数生成规则段落。
+        """
+        base = _SYSTEM_PROMPT_BASE.strip()
+
+        if global_constraint and global_constraint.has_param_directive:
+            directive_section = self._build_param_directive_section(
+                global_constraint
+            )
+            return base + "\n" + directive_section.strip()
+
+        return base
+
+    def _build_param_directive_section(
+        self,
+        constraint: ParamConstraint,
+    ) -> str:
+        """构建参数生成规则段落，注入 System Prompt。"""
+        lines: list[str] = []
+
+        if constraint.min_groups > 2:
+            lines.append(
+                f"- 每个接口的测试用例总数不少于 **{constraint.min_groups} 组**"
+                f"（正向 + 负向合计）"
+            )
+
+        strategy_labels = {
+            ParamStrategy.BOUNDARY:    "边界值（空值、最大值、最小值、临界值）",
+            ParamStrategy.EXCEPTION:   "异常值（None、类型错误、超长、特殊字符等）",
+            ParamStrategy.EQUIVALENCE: "等价类（有效等价类 + 无效等价类）",
+            ParamStrategy.RANDOM:      "随机值（覆盖典型随机场景）",
+        }
+        if constraint.strategies:
+            strategy_str = "、".join(
+                strategy_labels[s]
+                for s in constraint.strategies
+                if s in strategy_labels
+            )
+            lines.append(f"- 参数值必须覆盖以下类型：{strategy_str}")
+
+        if not lines:
+            return ""
+
+        directives = "\n  ".join(lines)
+        return _PARAM_DIRECTIVE_TEMPLATE.format(directives=directives)
+
+    # ── 项目信息头 ───────────────────────────────────────────
+
+    def build_project_header(
+        self,
+        project:      str = "",
+        project_desc: str = "",
+    ) -> str:
+        if not project and not project_desc:
+            return ""
+        lines = ["## 项目信息"]
+        if project:
+            lines.append(f"项目名称：{project}")
+        if project_desc:
+            lines.append(f"项目描述：{project_desc}")
+        return "\n".join(lines)
+
+    # ── 参数数据集段落 ───────────────────────────────────────
+
+    def build_param_dataset_section(
+        self,
+        iface_summaries: list[dict],
+        constraint:      ParamConstraint,
+    ) -> str:
+        """
+        为批次内所有接口生成参数数据集，返回注入 user_prompt 的文本段落。
+        """
+        if not constraint.has_param_directive:
+            return ""
+
+        all_lines: list[str] = [
+            "## 参数测试数据集",
+            f"> 以下数据集由系统自动生成，请从中选取参数值组合，"
+            f"确保每个接口的测试用例总数不少于 **{constraint.min_groups} 组**。",
+            "",
+        ]
+
+        for iface in iface_summaries:
+            name     = iface.get("name", "未知接口")
+            datasets = self._param_gen.generate_for_interface(
+                interface_summary=iface,
+                strategies=constraint.strategies,
+                min_groups=constraint.min_groups,
+            )
+            if not datasets:
+                continue
+
+            all_lines.append(f"### 接口：{name}")
+            text = self._param_gen.to_prompt_text(datasets, constraint.min_groups)
+            all_lines.append(text)
+            all_lines.append("")
+
+        return "\n".join(all_lines)
+
+    # ── 分批 user_prompt ────────────────────────────────────
+
+    def build_batch_prompt(
+        self,
+        batch:             list[dict],
+        requirements:      list[str],
+        project_header:    str = "",
+        param_constraint:  ParamConstraint = None,
+    ) -> str:
+        """
+        构建单批次的 user_prompt。
+        若存在参数约束，自动注入参数数据集。
+        """
+        req_lines = "\n".join(
+            f"{i + 1}. {r}" for i, r in enumerate(requirements)
+        )
+        parts: list[str] = []
+
+        if project_header:
+            parts.append(project_header)
+
+        parts.append(
+            "## 本批次接口描述\n"
+            + json.dumps(batch, ensure_ascii=False, indent=2)
+        )
+        parts.append("## 测试需求\n" + req_lines)
+
+        # 注入参数数据集
+        if param_constraint and param_constraint.has_param_directive:
+            dataset_section = self.build_param_dataset_section(
+                iface_summaries=batch,
+                constraint=param_constraint,
+            )
+            if dataset_section:
+                parts.append(dataset_section)
+
+        parts.append(
+            "请根据以上接口描述、测试需求和参数数据集，生成完整的测试用例（包含正向和负向）。\n"
+            "- function 协议：使用 module_path 字段导入被测函数\n"
+            "- http 协议：使用 full_url 字段作为请求地址\n"
+            "- 每条需求的 requirement_id 须填入对应用例的 requirement_id 字段\n"
+            "- test_data_notes 须说明本用例使用的参数类别和具体值"
+        )
+        return "\n\n".join(parts)
+
+    # ── 单次 user_prompt ────────────────────────────────────
 
     def build_user_prompt(
         self,
-        requirements:    list[str],
-        interfaces:      list[InterfaceInfo],
-        parser:          InterfaceParser,
-        project:         str = "",
-        project_desc:    str = "",
+        requirements:     list[str],
+        interfaces:       list[InterfaceInfo],
+        parser:           InterfaceParser,
+        project:          str = "",
+        project_desc:     str = "",
+        param_constraint: ParamConstraint = None,
     ) -> str:
+        """
+        单次调用模式（接口数 ≤ batch_size 时使用）。
+        """
         iface_summary = parser.to_summary_dict(interfaces)
         req_lines = "\n".join(
             f"{i + 1}. {r}" for i, r in enumerate(requirements)
         )
+        header = self.build_project_header(project, project_desc)
 
-        # 项目信息头
-        project_section = ""
-        if project or project_desc:
-            project_section = "## Project\n"
-            if project:
-                project_section += f"Name       : {project}\n"
-            if project_desc:
-                project_section += f"Description: {project_desc}\n"
-            project_section += "\n"
+        parts: list[str] = []
+        if header:
+            parts.append(header)
+        parts.append(
+            "## 接口描述\n"
+            + json.dumps(iface_summary, ensure_ascii=False, indent=2)
+        )
+        parts.append("## 测试需求\n" + req_lines)
 
-        return (
-            f"{project_section}"
-            f"## Available Interfaces\n"
-            f"{json.dumps(iface_summary, ensure_ascii=False, indent=2)}\n\n"
-            f"## Test Requirements\n"
-            f"Generate comprehensive test cases (positive + negative) for every interface.\n"
-            f"- The generated test cases for each interface must meet the following requirements: {req_lines}\n"
-            f"- For function interfaces: use 'module_path' for import, "
-            f"'source_file' for reference.\n"
-            f"- For HTTP interfaces: use 'full_url' as the request URL.\n"
-            f"- Use requirement_id from the interface to populate the test case's "
-            f"requirement_id field.\n"
-            f"- Chain multiple interface calls in one test case when a requirement "
-            f"involves more than one interface.\n"
-        ).strip()
\ No newline at end of file
+        if param_constraint and param_constraint.has_param_directive:
+            dataset_section = self.build_param_dataset_section(
+                iface_summaries=iface_summary
+                if isinstance(iface_summary, list) else list(iface_summary.values()),
+                constraint=param_constraint,
+            )
+            if dataset_section:
+                parts.append(dataset_section)
+
+        parts.append(
+            "请根据以上接口描述、测试需求和参数数据集，生成完整的测试用例（包含正向和负向）。\n"
+            "- function 协议：使用 module_path 字段导入被测函数\n"
+            "- http 协议：使用 full_url 字段作为请求地址\n"
+            "- 每条需求的 requirement_id 须填入对应用例的 requirement_id 字段\n"
+            "- test_data_notes 须说明本用例使用的参数类别和具体值"
+        )
+        return "\n\n".join(parts)
\ No newline at end of file
diff --git a/ai_test_generator/core/requirement_parser.py b/ai_test_generator/core/requirement_parser.py
new file mode 100644
index 0000000..7cce453
--- /dev/null
+++ b/ai_test_generator/core/requirement_parser.py
@@ -0,0 +1,211 @@
+"""
+需求解析器
+══════════════════════════════════════════════════════════════
+职责：
+  从用户填写的测试需求文本中，识别并提取参数生成约束指令：
+    - 参数组数约束：如"不少于10组"、"至少5组"、"生成8组"
+    - 参数策略约束：如"边界值"、"异常值"、"等价类"、"随机值"
+    - 数据类型约束：如"字符串边界值"、"整数异常值"
+  解析结果供 ParamGenerator 和 PromptBuilder 使用。
+"""
+
+from __future__ import annotations
+import re
+from dataclasses import dataclass, field
+from enum import Enum
+
+
+# ══════════════════════════════════════════════════════════════
+# 枚举：参数生成策略
+# ══════════════════════════════════════════════════════════════
+
+class ParamStrategy(str, Enum):
+    BOUNDARY    = "boundary"     # 边界值
+    EXCEPTION   = "exception"    # 异常值 / 错误值
+    EQUIVALENCE = "equivalence"  # 等价类
+    RANDOM      = "random"       # 随机值
+    CUSTOM      = "custom"       # 自定义（用户在需求中直接指定）
+
+
+# ══════════════════════════════════════════════════════════════
+# 数据结构：解析结果
+# ══════════════════════════════════════════════════════════════
+
+@dataclass
+class ParamConstraint:
+    """
+    从单条需求文本中解析出的参数生成约束。
+    """
+    # 最少生成的参数组数（正向 + 负向合计）
+    min_groups: int = 2
+
+    # 需要覆盖的策略集合
+    strategies: list[ParamStrategy] = field(default_factory=list)
+
+    # 原始需求文本（去除参数指令后的纯业务需求部分）
+    clean_requirement: str = ""
+
+    # 是否显式指定了参数要求（用于区分"用户有要求"和"默认值"）
+    has_param_directive: bool = False
+
+    def __str__(self) -> str:
+        parts = []
+        if self.min_groups > 2:
+            parts.append(f"至少 {self.min_groups} 组参数")
+        for s in self.strategies:
+            parts.append(_STRATEGY_LABEL[s])
+        return "、".join(parts) if parts else "默认（1正1负）"
+
+
+_STRATEGY_LABEL: dict[ParamStrategy, str] = {
+    ParamStrategy.BOUNDARY:    "边界值",
+    ParamStrategy.EXCEPTION:   "异常值",
+    ParamStrategy.EQUIVALENCE: "等价类",
+    ParamStrategy.RANDOM:      "随机值",
+    ParamStrategy.CUSTOM:      "自定义",
+}
+
+
+# ══════════════════════════════════════════════════════════════
+# 解析规则
+# ══════════════════════════════════════════════════════════════
+
+# 数量约束：匹配"不少于N组"、"至少N组"、"最少N组"、"生成N组"、"N组以上"
+_RE_MIN_GROUPS = re.compile(
+    r'(?:不少于|至少|最少|生成|共|需要|要求)\s*(\d+)\s*组'
+    r'|(\d+)\s*组(?:以上|及以上|或以上)',
+    re.UNICODE,
+)
+
+# 策略关键词映射（关键词 → 策略枚举）
+_STRATEGY_KEYWORDS: list[tuple[list[str], ParamStrategy]] = [
+    (["边界值", "边界", "boundary", "边界测试"],          ParamStrategy.BOUNDARY),
+    (["异常值", "异常", "错误值", "非法值",
+      "exception", "invalid", "error"],                  ParamStrategy.EXCEPTION),
+    (["等价类", "等价", "equivalence"],                   ParamStrategy.EQUIVALENCE),
+    (["随机", "随机值", "random"],                        ParamStrategy.RANDOM),
+]
+
+# 参数指令整体识别：包含上述任意关键词则认为是参数指令
+_RE_PARAM_DIRECTIVE = re.compile(
+    r'(?:参数|测试参数|测试数据|数据|入参)'
+    r'.*?(?:不少于|至少|最少|生成|边界|异常|等价|随机|\d+\s*组)',
+    re.UNICODE,
+)
+
+
+# ══════════════════════════════════════════════════════════════
+# 解析器
+# ══════════════════════════════════════════════════════════════
+
+class RequirementParser:
+    """
+    解析测试需求列表，提取参数生成约束，返回：
+      - clean_requirements : 去除参数指令后的纯业务需求列表
+      - constraints        : 每条需求对应的 ParamConstraint
+      - global_constraint  : 全局约束（从"全局"/"所有接口"等描述中提取）
+    """
+
+    def parse(
+        self,
+        requirements: list[str],
+    ) -> tuple[list[str], list[ParamConstraint], ParamConstraint]:
+        """
+        Returns:
+          (clean_requirements, per_req_constraints, global_constraint)
+        """
+        clean_reqs:  list[str]             = []
+        constraints: list[ParamConstraint] = []
+        global_c     = ParamConstraint()
+
+        for req in requirements:
+            c = self._parse_one(req)
+            constraints.append(c)
+            clean_reqs.append(c.clean_requirement)
+
+            # 若某条需求是全局性描述（如"所有接口参数不少于10组"），
+            # 则将其约束提升为全局约束
+            if self._is_global(req) and c.has_param_directive:
+                global_c = self._merge(global_c, c)
+
+        return clean_reqs, constraints, global_c
+
+    # ── 解析单条需求 ──────────────────────────────────────────
+
+    def _parse_one(self, req: str) -> ParamConstraint:
+        c = ParamConstraint(clean_requirement=req)
+
+        # 1. 检测是否包含参数指令
+        if not self._has_directive(req):
+            return c
+
+        c.has_param_directive = True
+
+        # 2. 提取数量约束
+        m = _RE_MIN_GROUPS.search(req)
+        if m:
+            n = int(m.group(1) or m.group(2))
+            c.min_groups = max(n, 2)   # 至少保留 1正1负
+
+        # 3. 提取策略约束
+        seen: set[ParamStrategy] = set()
+        for keywords, strategy in _STRATEGY_KEYWORDS:
+            if any(kw in req for kw in keywords):
+                seen.add(strategy)
+        c.strategies = list(seen)
+
+        # 4. 若未指定策略，默认同时覆盖边界值和异常值
+        if not c.strategies:
+            c.strategies = [ParamStrategy.BOUNDARY, ParamStrategy.EXCEPTION]
+
+        # 5. 清理需求文本：去除参数指令部分，保留业务描述
+        c.clean_requirement = self._strip_directive(req)
+
+        return c
+
+    # ── 工具方法 ──────────────────────────────────────────────
+
+    @staticmethod
+    def _has_directive(req: str) -> bool:
+        """判断需求文本是否包含参数生成指令。"""
+        return bool(_RE_PARAM_DIRECTIVE.search(req)) or bool(
+            _RE_MIN_GROUPS.search(req)
+        )
+
+    @staticmethod
+    def _is_global(req: str) -> bool:
+        """判断是否为全局性参数要求。"""
+        return any(
+            kw in req
+            for kw in ["所有接口", "全部接口", "全局", "所有测试", "全部测试"]
+        )
+
+    @staticmethod
+    def _strip_directive(req: str) -> str:
+        """
+        去除需求文本中的参数指令部分，保留纯业务描述。
+        策略：以中文逗号、分号、括号等分隔，去除含参数指令的子句。
+        """
+        # 按常见分隔符拆分
+        parts = re.split(r'[，,；;（(）)【】\n]', req)
+        clean_parts = []
+        for part in parts:
+            part = part.strip()
+            if not part:
+                continue
+            # 若该子句包含参数指令关键词，则跳过
+            if _RE_PARAM_DIRECTIVE.search(part) or _RE_MIN_GROUPS.search(part):
+                continue
+            clean_parts.append(part)
+        result = "，".join(clean_parts).strip()
+        return result if result else req   # 若全部被去除则保留原文
+
+    @staticmethod
+    def _merge(base: ParamConstraint, other: ParamConstraint) -> ParamConstraint:
+        """合并两个约束，取较大值。"""
+        merged = ParamConstraint(
+            min_groups=max(base.min_groups, other.min_groups),
+            strategies=list(set(base.strategies) | set(other.strategies)),
+            has_param_directive=True,
+        )
+        return merged
\ No newline at end of file
diff --git a/ai_test_generator/core/test_runner.py b/ai_test_generator/core/test_runner.py
index 90a3e7f..0f91940 100644
--- a/ai_test_generator/core/test_runner.py
+++ b/ai_test_generator/core/test_runner.py
@@ -1,168 +1,278 @@
-import os
-import sys
+"""
+测试执行器
+══════════════════════════════════════════════════════════════
+大规模支持改造：
+  1. 并行执行：ThreadPoolExecutor，默认 8 个并发
+  2. 超时控制：单个用例超时不阻塞整体
+  3. 步骤级结果解析：##STEP_RESULT## 行
+  4. 进度显示：实时打印完成数
+  5. 结果持久化：保存 run_results.json
+"""
+
+from __future__ import annotations
 import json
-import subprocess
-import time
 import logging
-from pathlib import Path
+import subprocess
+import sys
+import time
+from concurrent.futures import ThreadPoolExecutor, as_completed
 from dataclasses import dataclass, field
+from pathlib import Path
+
 from config import config
 
 logger = logging.getLogger(__name__)
 
+STEP_RESULT_PREFIX = "##STEP_RESULT##"
+
+
+# ══════════════════════════════════════════════════════════════
+# 数据结构
+# ══════════════════════════════════════════════════════════════
+
+@dataclass
+class AssertionResult:
+    field:    str
+    operator: str
+    expected: object
+    actual:   object
+    passed:   bool
+    message:  str = ""
+
 
 @dataclass
 class StepResult:
     step_no:        int
     interface_name: str
-    status:         str        # PASS | FAIL | ERROR
-    assertions:     list[dict] = field(default_factory=list)
-    # 每条 assertion: {"field","operator","expected","actual","passed","message"}
+    status:         str                          # "PASS" | "FAIL"
+    assertions:     list[AssertionResult] = field(default_factory=list)
 
 
 @dataclass
 class TestResult:
     test_id:      str
     file_path:    str
-    status:       str          # PASS | FAIL | ERROR | TIMEOUT
-    message:      str  = ""
-    duration:     float = 0.0
-    stdout:       str  = ""
-    stderr:       str  = ""
+    status:       str                            # "PASS" | "FAIL" | "ERROR" | "TIMEOUT"
+    message:      str        = ""
+    duration:     float      = 0.0
     step_results: list[StepResult] = field(default_factory=list)
+    stdout:       str        = ""
+    stderr:       str        = ""
 
 
+# ══════════════════════════════════════════════════════════════
+# 执行器
+# ══════════════════════════════════════════════════════════════
+
 class TestRunner:
 
+    def __init__(self):
+        self.timeout     = getattr(config, "TEST_TIMEOUT",     60)
+        self.max_workers = getattr(config, "TEST_MAX_WORKERS",  8)
+        self.python_bin  = getattr(config, "TEST_PYTHON_BIN",  sys.executable)
+
+    # ── 主入口 ────────────────────────────────────────────────
+
     def run_all(self, test_files: list[Path]) -> list[TestResult]:
-        results = []
-        total = len(test_files)
-        print(f"\n{'═'*62}")
-        print(f"  Running {total} test(s) …")
-        print(f"{'═'*62}")
-        for idx, fp in enumerate(test_files, 1):
-            print(f"\n[{idx}/{total}] {fp.name}")
-            result = self._run_one(fp)
-            results.append(result)
-            self._print_result(result)
+        """
+        并行执行所有测试文件，返回结果列表。
+        """
+        if not test_files:
+            logger.warning("No test files to run.")
+            return []
+
+        total   = len(test_files)
+        results: list[TestResult] = []
+        done    = 0
+
+        logger.info(
+            f"Running {total} test(s) with "
+            f"max_workers={self.max_workers}, timeout={self.timeout}s"
+        )
+
+        with ThreadPoolExecutor(max_workers=self.max_workers) as executor:
+            future_map = {
+                executor.submit(self._run_one, f): f
+                for f in test_files
+            }
+            for future in as_completed(future_map):
+                result = future.result()
+                results.append(result)
+                done += 1
+                icon = "✅" if result.status == "PASS" else "❌"
+                logger.info(
+                    f"  [{done}/{total}] {icon} {result.test_id} "
+                    f"({result.status}) {result.duration:.2f}s"
+                )
+
+        # 按文件名排序，保持输出稳定
+        results.sort(key=lambda r: r.file_path)
         return results
 
-    # ── 执行单个脚本 ──────────────────────────────────────────
+    # ── 执行单个测试文件 ──────────────────────────────────────
 
     def _run_one(self, file_path: Path) -> TestResult:
-        test_id = file_path.stem
-        t0 = time.time()
+        test_id   = file_path.stem
+        start     = time.monotonic()
+
         try:
             proc = subprocess.run(
-                [sys.executable, str(file_path)],
-                capture_output=True, text=True,
-                timeout=config.TEST_TIMEOUT,
-                env=self._env(),
+                [self.python_bin, str(file_path)],
+                capture_output=True,
+                text=True,
+                timeout=self.timeout,
             )
-            duration = time.time() - t0
-            stdout   = proc.stdout.strip()
-            stderr   = proc.stderr.strip()
-            status, message = self._parse_output(stdout, proc.returncode)
-            step_results    = self._parse_step_results(stdout)
-            return TestResult(
-                test_id=test_id, file_path=str(file_path),
-                status=status, message=message,
-                duration=duration, stdout=stdout, stderr=stderr,
-                step_results=step_results,
+            duration = time.monotonic() - start
+            return self._parse_output(
+                test_id, str(file_path), proc, duration
             )
+
         except subprocess.TimeoutExpired:
+            duration = time.monotonic() - start
+            logger.warning(f"  TIMEOUT: {test_id} ({duration:.1f}s)")
             return TestResult(
-                test_id=test_id, file_path=str(file_path),
-                status="TIMEOUT", message=f"Exceeded {config.TEST_TIMEOUT}s",
-                duration=time.time() - t0,
+                test_id=test_id,
+                file_path=str(file_path),
+                status="TIMEOUT",
+                message=f"Exceeded {self.timeout}s timeout",
+                duration=duration,
             )
         except Exception as e:
+            duration = time.monotonic() - start
+            logger.error(f"  ERROR running {test_id}: {e}")
             return TestResult(
-                test_id=test_id, file_path=str(file_path),
-                status="ERROR", message=str(e),
-                duration=time.time() - t0,
+                test_id=test_id,
+                file_path=str(file_path),
+                status="ERROR",
+                message=str(e),
+                duration=duration,
             )
 
     # ── 解析输出 ──────────────────────────────────────────────
 
-    def _parse_output(self, stdout: str, returncode: int) -> tuple[str, str]:
-        if not stdout:
-            return "FAIL", f"No output (exit={returncode})"
-        last = stdout.strip().splitlines()[-1].strip()
-        upper = last.upper()
-        if upper.startswith("PASS"):
-            return "PASS", last[5:].strip()
-        if upper.startswith("FAIL"):
-            return "FAIL", last[5:].strip()
-        return ("PASS" if returncode == 0 else "FAIL"), last
+    def _parse_output(
+        self,
+        test_id:   str,
+        file_path: str,
+        proc:      subprocess.CompletedProcess,
+        duration:  float,
+    ) -> TestResult:
+        stdout       = proc.stdout or ""
+        stderr       = proc.stderr or ""
+        step_results = self._parse_step_results(stdout)
+
+        # 最后一行决定整体状态
+        last_line = stdout.strip().splitlines()[-1] if stdout.strip() else ""
+        if last_line.startswith("PASS"):
+            status  = "PASS"
+            message = last_line
+        elif last_line.startswith("FAIL"):
+            status  = "FAIL"
+            message = last_line
+        elif proc.returncode != 0:
+            status  = "FAIL"
+            message = f"Exit code {proc.returncode}: {stderr[:200]}"
+        else:
+            status  = "ERROR"
+            message = "No PASS/FAIL line found in output"
+
+        return TestResult(
+            test_id=test_id,
+            file_path=file_path,
+            status=status,
+            message=message,
+            duration=duration,
+            step_results=step_results,
+            stdout=stdout,
+            stderr=stderr,
+        )
 
     def _parse_step_results(self, stdout: str) -> list[StepResult]:
-        """
-        解析脚本中以 ##STEP_RESULT## 开头的结构化输出行
-        格式：##STEP_RESULT## <json>
-        """
         results = []
         for line in stdout.splitlines():
             line = line.strip()
-            if line.startswith("##STEP_RESULT##"):
-                try:
-                    data = json.loads(line[len("##STEP_RESULT##"):].strip())
-                    results.append(StepResult(
-                        step_no=data.get("step_no", 0),
-                        interface_name=data.get("interface_name", ""),
-                        status=data.get("status", ""),
-                        assertions=data.get("assertions", []),
-                    ))
-                except Exception:
-                    pass
+            if not line.startswith(STEP_RESULT_PREFIX):
+                continue
+            try:
+                raw = json.loads(line[len(STEP_RESULT_PREFIX):].strip())
+                assertions = [
+                    AssertionResult(
+                        field=a.get("field", ""),
+                        operator=a.get("operator", ""),
+                        expected=a.get("expected"),
+                        actual=a.get("actual"),
+                        passed=bool(a.get("passed", False)),
+                        message=a.get("message", ""),
+                    )
+                    for a in raw.get("assertions", [])
+                ]
+                results.append(StepResult(
+                    step_no=raw.get("step_no", 0),
+                    interface_name=raw.get("interface_name", ""),
+                    status=raw.get("status", "FAIL"),
+                    assertions=assertions,
+                ))
+            except (json.JSONDecodeError, KeyError) as e:
+                logger.debug(f"Step result parse error: {e}")
         return results
 
-    def _env(self) -> dict:
-        env = os.environ.copy()
-        env["HTTP_BASE_URL"] = config.HTTP_BASE_URL
-        return env
-
-    # ── 打印 ──────────────────────────────────────────────────
-
-    def _print_result(self, r: TestResult):
-        icon = {"PASS": "✅", "FAIL": "❌", "TIMEOUT": "⏱️", "ERROR": "⚠️"}.get(r.status, "❓")
-        print(f"  {icon} [{r.status}] {r.test_id}  ({r.duration:.2f}s)")
-        print(f"     {r.message}")
-        for sr in r.step_results:
-            s_icon = "✅" if sr.status == "PASS" else "❌"
-            print(f"     {s_icon} Step {sr.step_no}: {sr.interface_name}")
-            for a in sr.assertions:
-                a_icon = "✔" if a.get("passed") else "✘"
-                print(f"        {a_icon} {a.get('message','')} "
-                      f"(expected={a.get('expected')}, actual={a.get('actual')})")
-        if r.stderr:
-            print(f"     stderr: {r.stderr[:300]}")
+    # ── 摘要打印 ──────────────────────────────────────────────
 
     def print_summary(self, results: list[TestResult]):
-        total  = len(results)
-        passed = sum(1 for r in results if r.status == "PASS")
-        failed = sum(1 for r in results if r.status == "FAIL")
-        errors = sum(1 for r in results if r.status in ("ERROR", "TIMEOUT"))
-        print(f"\n{'═'*62}")
-        print(f"  TEST SUMMARY")
-        print(f"{'═'*62}")
-        print(f"  Total    : {total}")
+        total   = len(results)
+        passed  = sum(1 for r in results if r.status == "PASS")
+        failed  = sum(1 for r in results if r.status == "FAIL")
+        errors  = sum(1 for r in results if r.status in ("ERROR", "TIMEOUT"))
+        avg_dur = sum(r.duration for r in results) / total if total else 0
+
+        print(f"\n{'─' * 56}")
+        print(f"  Test Run Summary  ({total} cases)")
+        print(f"{'─' * 56}")
         print(f"  ✅ PASS   : {passed}")
         print(f"  ❌ FAIL   : {failed}")
         print(f"  ⚠️  ERROR  : {errors}")
-        print(f"  Pass Rate: {passed/total*100:.1f}%" if total else "  Pass Rate: N/A")
-        print(f"{'═'*62}\n")
+        print(f"  ⏱  Avg    : {avg_dur:.2f}s / case")
+        print(f"{'─' * 56}")
+
+        if failed or errors:
+            print("  Failed / Error cases:")
+            for r in results:
+                if r.status not in ("PASS",):
+                    print(f"    [{r.status}] {r.test_id}: {r.message}")
+        print()
+
+    # ── 持久化 ────────────────────────────────────────────────
 
     def save_results(self, results: list[TestResult], path: str):
-        with open(path, "w", encoding="utf-8") as f:
-            json.dump([{
-                "test_id": r.test_id, "status": r.status,
-                "message": r.message, "duration": r.duration,
-                "stdout": r.stdout, "stderr": r.stderr,
+        data = [
+            {
+                "test_id":    r.test_id,
+                "file_path":  r.file_path,
+                "status":     r.status,
+                "message":    r.message,
+                "duration":   round(r.duration, 3),
                 "step_results": [
-                    {"step_no": sr.step_no, "interface_name": sr.interface_name,
-                     "status": sr.status, "assertions": sr.assertions}
-                    for sr in r.step_results
+                    {
+                        "step_no":        s.step_no,
+                        "interface_name": s.interface_name,
+                        "status":         s.status,
+                        "assertions": [
+                            {
+                                "field":    a.field,
+                                "operator": a.operator,
+                                "expected": a.expected,
+                                "actual":   a.actual,
+                                "passed":   a.passed,
+                                "message":  a.message,
+                            }
+                            for a in s.assertions
+                        ],
+                    }
+                    for s in r.step_results
                 ],
-            } for r in results], f, ensure_ascii=False, indent=2)
+            }
+            for r in results
+        ]
+        with open(path, "w", encoding="utf-8") as f:
+            json.dump(data, f, ensure_ascii=False, indent=2)
         logger.info(f"Run results → {path}")
\ No newline at end of file
diff --git a/ai_test_generator/main.py b/ai_test_generator/main.py
index 5714c79..7ff44e7 100644
--- a/ai_test_generator/main.py
+++ b/ai_test_generator/main.py
@@ -3,11 +3,18 @@
 AI-Powered API Test Generator, Runner & Coverage Analyzer
 ──────────────────────────────────────────────────────────
 Usage:
+  # 基础用法
   python main.py --api-desc examples/api_desc.json \\
                  --requirements "创建用户,删除用户"
 
+  # 在需求中自然语言描述参数生成要求
   python main.py --api-desc examples/api_desc.json \\
-                 --req-file  examples/requirements.txt
+                 --requirements "创建用户（测试参数不少于10组，覆盖边界值和异常值）,删除用户"
+
+  # 从文件读取需求（每行一条，支持参数生成指令）
+  python main.py --api-desc examples/api_desc.json \\
+                 --req-file  examples/requirements.txt \\
+                 --batch-size 10
 """
 
 import argparse
@@ -17,6 +24,7 @@ from pathlib import Path
 
 from config import config
 from core.parser import InterfaceParser, ApiDescriptor
+from core.requirement_parser import RequirementParser, ParamConstraint
 from core.prompt_builder import PromptBuilder
 from core.llm_client import LLMClient
 from core.test_generator import TestGenerator
@@ -24,48 +32,57 @@ from core.test_runner import TestRunner
 from core.analyzer import CoverageAnalyzer, CoverageReport
 from core.report_generator import ReportGenerator
 
-logging.basicConfig(
-    level=logging.INFO,
-    format="%(asctime)s [%(levelname)s] %(name)s — %(message)s",
-    handlers=[logging.StreamHandler(sys.stdout)],
-)
+
+# ══════════════════════════════════════════════════════════════
+# 日志初始化
+# ══════════════════════════════════════════════════════════════
+
+def setup_logging(debug: bool = False):
+    """
+    配置日志级别。
+
+    ⚠️  Bug 规避（openai-python v2.24.0 + pydantic-core）：
+        第三方 SDK logger 强制锁定在 WARNING，避免触发
+        model_dump(by_alias=None) → pydantic-core Rust 层 TypeError。
+    Ref: https://github.com/openai/openai-python/issues/2921
+    """
+    root_level = logging.DEBUG if debug else logging.INFO
+    logging.basicConfig(
+        level=root_level,
+        format="%(asctime)s [%(levelname)s] %(name)s — %(message)s",
+        handlers=[logging.StreamHandler(sys.stdout)],
+    )
+    for name in ["openai", "openai._base_client", "anthropic", "httpx", "httpcore"]:
+        logging.getLogger(name).setLevel(logging.WARNING)
+
+
 logger = logging.getLogger(__name__)
 
 
-# ── CLI ───────────────────────────────────────────────────────
+# ══════════════════════════════════════════════════════════════
+# CLI
+# ══════════════════════════════════════════════════════════════
 
 def build_arg_parser() -> argparse.ArgumentParser:
     p = argparse.ArgumentParser(
-        description="AI-Powered API Test Generator & Coverage Analyzer"
-    )
-    p.add_argument(
-        "--api-desc", required=True,
-        help="接口描述 JSON 文件（含 project / description / units 字段）",
-    )
-    p.add_argument(
-        "--requirements", default="",
-        help="测试需求（逗号分隔）",
-    )
-    p.add_argument(
-        "--req-file", default="",
-        help="测试需求文件（每行一条）",
-    )
-    p.add_argument(
-        "--skip-run", action="store_true",
-        help="只生成测试文件，不执行",
-    )
-    p.add_argument(
-        "--skip-analyze", action="store_true",
-        help="跳过覆盖率分析",
-    )
-    p.add_argument(
-        "--output-dir", default="",
-        help="测试文件根输出目录（默认 config.GENERATED_TESTS_DIR）",
-    )
-    p.add_argument(
-        "--debug", action="store_true",
-        help="开启 DEBUG 日志",
+        description="AI-Powered API Test Generator & Coverage Analyzer",
+        formatter_class=argparse.RawDescriptionHelpFormatter,
+        epilog="""
+参数生成指令示例（可直接写在需求文本中）：
+  "创建用户，测试参数不少于10组，覆盖边界值和异常值"
+  "查询接口（至少8组，边界值）"
+  "所有接口参数不少于5组，覆盖等价类"
+        """,
     )
+    p.add_argument("--api-desc",     required=True,       help="接口描述 JSON 文件")
+    p.add_argument("--requirements", default="",          help="测试需求（逗号分隔，支持参数生成指令）")
+    p.add_argument("--req-file",     default="",          help="测试需求文件（每行一条）")
+    p.add_argument("--batch-size",   type=int, default=0, help="每批接口数量（0=使用 config.LLM_BATCH_SIZE）")
+    p.add_argument("--workers",      type=int, default=0, help="并行执行线程数（0=使用 config.TEST_MAX_WORKERS）")
+    p.add_argument("--skip-run",     action="store_true", help="只生成，不执行")
+    p.add_argument("--skip-analyze", action="store_true", help="跳过覆盖率分析")
+    p.add_argument("--output-dir",   default="",          help="测试文件根输出目录")
+    p.add_argument("--debug",        action="store_true", help="开启 DEBUG 日志（第三方 SDK 仍保持 WARNING）")
     return p
 
 
@@ -77,82 +94,111 @@ def load_requirements(args) -> list[str]:
         with open(args.req_file, "r", encoding="utf-8") as f:
             reqs += [line.strip() for line in f if line.strip()]
     if not reqs:
-        logger.error(
-            "No requirements provided. Use --requirements or --req-file."
-        )
+        logger.error("未提供任何测试需求，请使用 --requirements 或 --req-file 指定。")
         sys.exit(1)
     return reqs
 
 
-# ── Main ──────────────────────────────────────────────────────
+# ══════════════════════════════════════════════════════════════
+# Main
+# ══════════════════════════════════════════════════════════════
 
 def main():
     args = build_arg_parser().parse_args()
+    setup_logging(debug=args.debug)
 
-    if args.debug:
-        logging.getLogger().setLevel(logging.DEBUG)
-    if args.output_dir:
-        config.GENERATED_TESTS_DIR = args.output_dir
+    if args.output_dir:    config.GENERATED_TESTS_DIR = args.output_dir
+    if args.batch_size > 0: config.LLM_BATCH_SIZE     = args.batch_size
+    if args.workers > 0:   config.TEST_MAX_WORKERS    = args.workers
 
     # ── Step 1: 解析接口描述 ──────────────────────────────────
-    logger.info("▶ Step 1: Parsing interface description …")
-    parser: InterfaceParser  = InterfaceParser()
-    descriptor: ApiDescriptor = parser.parse_file(args.api_desc)
+    logger.info("▶ Step 1: 解析接口描述 …")
+    parser     = InterfaceParser()
+    descriptor = parser.parse_file(args.api_desc)
 
     project      = descriptor.project
     project_desc = descriptor.description
     interfaces   = descriptor.interfaces
 
-    logger.info(f"  Project    : {project}")
-    logger.info(f"  Description: {project_desc}")
-    logger.info(
-        f"  Interfaces : {len(interfaces)} — "
-        f"{[i.name for i in interfaces]}"
-    )
-    # 打印每个接口的 url 解析结果
+    logger.info(f"  项目名称：{project}")
+    logger.info(f"  项目描述：{project_desc}")
+    logger.info(f"  接口数量：{len(interfaces)}")
     for iface in interfaces:
         if iface.protocol == "function":
             logger.info(
-                f"    [{iface.name}] source={iface.source_file} "
-                f"module={iface.module_path}"
+                f"    [function] {iface.name} "
+                f"← {iface.source_file} (module: {iface.module_path})"
             )
         else:
-            logger.info(
-                f"    [{iface.name}] full_url={iface.http_full_url}"
-            )
+            logger.info(f"    [http]     {iface.name} ← {iface.http_full_url}")
 
-    # ── Step 2: 加载测试需求 ──────────────────────────────────
-    logger.info("▶ Step 2: Loading test requirements …")
-    requirements = load_requirements(args)
-    for i, req in enumerate(requirements, 1):
-        logger.info(f"  {i}. {req}")
+    # ── Step 2: 加载并解析测试需求（含参数生成指令）──────────
+    logger.info("▶ Step 2: 加载并解析测试需求 …")
+    raw_requirements = load_requirements(args)
 
-    # ── Step 3: 调用 LLM 生成测试用例 ────────────────────────
-    logger.info("▶ Step 3: Calling LLM …")
-    builder    = PromptBuilder()
-    test_cases = LLMClient().generate_test_cases(
-        builder.get_system_prompt(),
-        builder.build_user_prompt(
-            requirements, interfaces, parser,
-            project=project,
-            project_desc=project_desc,
-        ),
+    req_parser = RequirementParser()
+    clean_reqs, per_req_constraints, global_constraint = req_parser.parse(
+        raw_requirements
     )
-    logger.info(f"  LLM returned {len(test_cases)} test case(s)")
+
+    # 打印原始需求 & 解析结果
+    for i, (raw, clean, c) in enumerate(
+        zip(raw_requirements, clean_reqs, per_req_constraints), 1
+    ):
+        if c.has_param_directive:
+            logger.info(
+                f"  {i}. {raw}\n"
+                f"     └─ 参数约束：{c}  |  业务需求：{clean}"
+            )
+        else:
+            logger.info(f"  {i}. {raw}")
+
+    if global_constraint.has_param_directive:
+        logger.info(f"  全局参数约束：{global_constraint}")
+
+    # 使用"清洗后的需求"传给 LLM（去掉参数指令，保留纯业务描述）
+    # 同时保留原始需求用于覆盖率分析（保证与用户输入一致）
+    effective_requirements = clean_reqs
+
+    # ── Step 3: 构建 Prompt 并调用 LLM 生成测试用例 ──────────
+    logger.info(
+        f"▶ Step 3: 调用 LLM 生成测试用例 "
+        f"(batch_size={getattr(config, 'LLM_BATCH_SIZE', 10)}) …"
+    )
+    builder        = PromptBuilder()
+    iface_summary  = parser.to_summary_dict(interfaces)
+    project_header = builder.build_project_header(project, project_desc)
+
+    # System Prompt：注入全局参数约束规则
+    system_prompt = builder.get_system_prompt(
+        global_constraint=global_constraint
+    )
+
+    test_cases = LLMClient().generate_test_cases(
+        system_prompt=system_prompt,
+        user_prompt="",
+        iface_summaries=iface_summary,
+        requirements=effective_requirements,
+        project_header=project_header,
+        param_constraint=global_constraint,   # ← 注入参数约束
+    )
+    logger.info(f"  共生成测试用例：{len(test_cases)} 个")
 
     # ── Step 4: 生成测试文件 ──────────────────────────────────
-    logger.info(
-        f"▶ Step 4: Generating test files (project='{project}') …"
-    )
+    logger.info("▶ Step 4: 生成测试文件 …")
     generator  = TestGenerator(project=project, project_desc=project_desc)
     test_files = generator.generate(test_cases)
     out        = generator.output_dir
+    logger.info(f"  输出目录：{out.resolve()}")
 
     run_results = []
 
     if not args.skip_run:
-        # ── Step 5: 执行测试 ──────────────────────────────────
-        logger.info("▶ Step 5: Running tests …")
+        # ── Step 5: 并行执行测试 ──────────────────────────────
+        logger.info(
+            f"▶ Step 5: 执行测试 "
+            f"(workers={getattr(config, 'TEST_MAX_WORKERS', 8)}) …"
+        )
         runner      = TestRunner()
         run_results = runner.run_all(test_files)
         runner.print_summary(run_results)
@@ -160,31 +206,36 @@ def main():
 
     if not args.skip_analyze:
         # ── Step 6: 覆盖率分析 ────────────────────────────────
-        logger.info("▶ Step 6: Analyzing coverage …")
+        logger.info("▶ Step 6: 覆盖率分析 …")
         report = CoverageAnalyzer(
             interfaces=interfaces,
-            requirements=requirements,
+            requirements=raw_requirements,   # 用原始需求做覆盖率分析
             test_cases=test_cases,
             run_results=run_results,
         ).analyze()
 
         # ── Step 7: 生成报告 ──────────────────────────────────
-        logger.info("▶ Step 7: Generating reports …")
+        logger.info("▶ Step 7: 生成报告 …")
         rg = ReportGenerator()
         rg.save_json(report, str(out / "coverage_report.json"))
         rg.save_html(report, str(out / "coverage_report.html"))
 
-        _print_terminal_summary(report, out, project)
+        _print_terminal_summary(report, out, project, global_constraint)
 
-    logger.info(f"\n✅ Done. Output: {out.resolve()}")
+    logger.info(f"\n✅ 完成。输出目录：{out.resolve()}")
 
 
-# ── 终端摘要 ──────────────────────────────────────────────────
+# ══════════════════════════════════════════════════════════════
+# 终端摘要
+# ══════════════════════════════════════════════════════════════
 
 def _print_terminal_summary(
-    report: CoverageReport, out: Path, project: str
+    report:            "CoverageReport",
+    out:               Path,
+    project:           str,
+    global_constraint: ParamConstraint,
 ):
-    W = 66
+    W = 68
 
     def bar(rate: float, w: int = 20) -> str:
         filled = int(rate * w)
@@ -193,35 +244,41 @@ def _print_terminal_summary(
         return f"{'█' * filled}{'░' * empty} {rate * 100:.1f}% {icon}"
 
     print(f"\n{'═' * W}")
-    print(f"  PROJECT : {project}")
-    print(f"  COVERAGE SUMMARY")
+    print(f"  项目：{project}")
+    print(f"  覆盖率摘要")
     print(f"{'═' * W}")
-    print(f"  接口覆盖率       {bar(report.interface_coverage_rate)}")
-    print(f"  需求覆盖率       {bar(report.requirement_coverage_rate)}")
-    print(f"  入参覆盖率       {bar(report.avg_in_param_coverage_rate)}")
-    print(f"  成功返回字段覆盖 {bar(report.avg_success_field_coverage_rate)}")
-    print(f"  失败返回字段覆盖 {bar(report.avg_failure_field_coverage_rate)}")
-    print(f"  用例通过率       {bar(report.pass_rate)}")
+    print(f"  接口覆盖率         {bar(report.interface_coverage_rate)}")
+    print(f"  需求覆盖率         {bar(report.requirement_coverage_rate)}")
+    print(f"  入参覆盖率         {bar(report.avg_in_param_coverage_rate)}")
+    print(f"  成功返回字段覆盖   {bar(report.avg_success_field_coverage_rate)}")
+    print(f"  失败返回字段覆盖   {bar(report.avg_failure_field_coverage_rate)}")
+    print(f"  用例通过率         {bar(report.pass_rate)}")
     print(f"{'─' * W}")
-    print(f"  Total Gaps : {len(report.gaps)}")
-    print(f"  🔴 Critical: {report.critical_gap_count}")
-    print(f"  🟠 High    : {report.high_gap_count}")
+    print(f"  测试用例总数       {report.total_test_cases}")
+    print(f"  覆盖缺口总数       {len(report.gaps)}")
+    print(f"  🔴 严重缺口        {report.critical_gap_count}")
+    print(f"  🟠 高优先级缺口    {report.high_gap_count}")
+
+    # 参数约束达成情况
+    if global_constraint.has_param_directive:
+        print(f"{'─' * W}")
+        print(f"  参数生成约束      {global_constraint}")
+        actual = report.total_test_cases
+        needed = global_constraint.min_groups
+        status = "✅ 已满足" if actual >= needed else f"❌ 未满足（需 {needed} 组，实际 {actual} 组）"
+        print(f"  参数组数要求      {status}")
 
     if report.gaps:
         print(f"{'─' * W}")
-        print("  Top Gaps (up to 8):")
-        icons = {
-            "critical": "🔴", "high": "🟠",
-            "medium":   "🟡", "low":  "🔵",
-        }
+        print("  Top 缺口（最多显示8条）：")
+        icons = {"critical": "🔴", "high": "🟠", "medium": "🟡", "low": "🔵"}
         for g in report.gaps[:8]:
-            icon = icons.get(g.severity, "⚪")
-            print(f"  {icon} [{g.gap_type}] {g.target}")
+            print(f"  {icons.get(g.severity, '⚪')} [{g.gap_type}] {g.target}")
             print(f"       → {g.suggestion}")
 
     print(f"{'─' * W}")
-    print(f"  Output : {out.resolve()}")
-    print(f"    • coverage_report.html  ← open in browser")
+    print(f"  输出目录：{out.resolve()}")
+    print(f"    • coverage_report.html")
     print(f"    • coverage_report.json")
     print(f"    • run_results.json")
     print(f"    • test_cases_summary.json")
diff --git a/ai_test_generator/run.sh b/ai_test_generator/run.sh
index 6da9f8c..c4d8f5e 100755
--- a/ai_test_generator/run.sh
+++ b/ai_test_generator/run.sh
@@ -16,4 +16,4 @@ export HTTP_BASE_URL="http://localhost:8080"
 
 # 只生成不执行
 python main.py --api-desc examples/api_desc.json \
-  --requirements "1.对每个接口进行测试，支持特殊值、边界值测试"
+  --requirements "每个测试用例的参数不少于10组"