diff --git a/pipeline_core/skill_loader.py b/pipeline_core/skill_loader.py
index 5ee78fb..01cf4d8 100644
--- a/pipeline_core/skill_loader.py
+++ b/pipeline_core/skill_loader.py
@@ -25,6 +25,7 @@ pipeline-core: Skill Loader — 六级可插拔技能隔离
 
 import logging
 import os
+import re
 from dataclasses import dataclass, field
 from typing import Dict, List, Optional
 
@@ -280,18 +281,29 @@ class SkillLoader:
                        role: str = "", project_id: str = "",
                        org_id: str = "", user_id: str = "",
                        max_skills: int = 5) -> List[Skill]:
-        """根据用户输入匹配相关技能（trigger_keywords + scope 加权）"""
+        """根据用户输入匹配相关技能（trigger_keywords + 技能名拆分 + scope 加权）。
+
+        兼容 Hermes 技能格式：多数 Hermes 技能只有 trigger_conditions（自然语言）而无
+        trigger_keywords，因此额外用技能名拆分匹配（module-development-spec → module/development/spec）
+        兜底，保证英文技能名也能被中文输入触达。
+        """
         all_skills = list(self.get_merged(pipeline_id, role, project_id, org_id, user_id).values())
         scored = []
         user_lower = user_input.lower()
         for skill in all_skills:
             score = 0
+            # 1. 显式 trigger_keywords（中文/英文关键词）
             for kw in skill.trigger_keywords:
                 if kw.lower() in user_lower:
-                    score += 1
+                    score += 3
+            # 2. 技能名整体匹配
             if skill.name.lower() in user_lower:
                 score += 2
-            # scope 优先级加权
+            # 3. 技能名拆分匹配（module-development-spec → module/development/spec）
+            for part in re.split(r'[-_]+', skill.name.lower()):
+                if len(part) >= 3 and part in user_lower:
+                    score += 1
+            # 4. scope 优先级加权
             score += SCOPE_PRIORITY.get(skill.scope, 0)
             if score > 0:
                 scored.append((score, skill))
@@ -322,10 +334,16 @@ class SkillLoader:
         if user_input:
             skills = self.get_by_trigger(user_input, pipeline_id, role,
                                          project_id, org_id, user_id, max_skills)
+            if not skills:
+                # 匹配不到时兜底：按 scope 优先级 + 名字稳定排序，避免目录层为空
+                merged = self.get_merged(pipeline_id, role, project_id, org_id, user_id)
+                skills = sorted(merged.values(),
+                                key=lambda s: (SCOPE_PRIORITY.get(s.scope, 9), s.name))
+                skills = skills[:max_skills]
         else:
             merged = self.get_merged(pipeline_id, role, project_id, org_id, user_id)
             skills = sorted(merged.values(),
-                            key=lambda s: SCOPE_PRIORITY.get(s.scope, 9))
+                            key=lambda s: (SCOPE_PRIORITY.get(s.scope, 9), s.name))
             skills = skills[:max_skills]
 
         if not skills: