feat: 添加web_scan的text_only参数支持Shadow DOM文本提取

assets/tools_schema.json

@@ -37,7 +37,8 @@
     "description": "获取当前页面的简化HTML内容和标签页列表。注意：简化会过滤边栏、浮动元素等非主体内容，如需查看被过滤内容请用execute_js。切换页面后一般应先调用查看。",
     "parameters": {"type": "object", "properties": {
       "tabs_only": {"type": "boolean", "description": "仅返回标签页列表和当前标签信息，不获取HTML内容。", "default": false},
-      "switch_tab_id": {"type": "string", "description": "可选的标签页 ID。如果提供，系统将在扫描前切换到该标签页。"}}}
+      "switch_tab_id": {"type": "string", "description": "可选的标签页 ID。如果提供，系统将在扫描前切换到该标签页。"},
+      "text_only": {"type": "boolean", "description": "只要纯文本不要HTML信息。", "default": false}}}
   }},
   {"type": "function", "function": {
     "name": "web_execute_js",

ga.py

@@ -109,7 +109,7 @@ def first_init_driver():
         #driver.newtab()
         time.sleep(3)
 
-def web_scan(tabs_only=False, switch_tab_id=None):
+def web_scan(tabs_only=False, switch_tab_id=None, text_only=False):
     """
     获取当前页面的简化HTML内容和标签页列表。注意：简化过程会过滤边栏、浮动元素等非主体内容。
     tabs_only: 仅返回标签页列表，不获取HTML内容（节省token）。
@@ -135,7 +135,7 @@ def web_scan(tabs_only=False, switch_tab_id=None):
                 "active_tab": driver.default_session_id
             }
         }
-        if not tabs_only: result["content"] = get_html(driver, cutlist=True, maxchars=28000)
+        if not tabs_only: result["content"] = get_html(driver, cutlist=True, maxchars=28000, text_only=text_only)
         return result
     except Exception as e:
         return {"status": "error", "msg": format_error(e)}
@@ -243,8 +243,7 @@ def smart_format(data, max_depth=2, max_str_len=100, omit_str=' ... '):
     return json.dumps(truncate(data, 0), indent=2, ensure_ascii=False, default=str)
 
 class GenericAgentHandler(BaseHandler):
-    '''Generic Agent 工具库，包含多种工具的实现。工具函数自动加上了 do_ 前缀。实际工具名没有前缀。
+    '''Generic Agent 工具库，包含多种工具的实现。工具函数自动加上了 do_ 前缀。实际工具名没有前缀。'''
-    '''
     def __init__(self, parent, last_history=None, cwd='./'):
         self.parent = parent
         self.key_info = ""
@@ -302,7 +301,8 @@ class GenericAgentHandler(BaseHandler):
         '''
         tabs_only = args.get("tabs_only", False)
         switch_tab_id = args.get("switch_tab_id", None)
-        result = web_scan(tabs_only=tabs_only, switch_tab_id=switch_tab_id)
+        text_only = args.get("text_only", False)
+        result = web_scan(tabs_only=tabs_only, switch_tab_id=switch_tab_id, text_only=text_only)
         content = result.pop("content", None)
         yield f'[Info] {str(result)}\n'
         if content: next_prompt = f"<tool_result>\n```html\n{content}\n```\n</tool_result>"
@@ -373,8 +373,7 @@ class GenericAgentHandler(BaseHandler):
                 old = open(path, 'r', encoding="utf-8").read() if os.path.exists(path) else ""
                 open(path, 'w', encoding="utf-8").write(new_content + old)
             else:
-                with open(path, 'a' if mode == "append" else 'w', encoding="utf-8") as f:
+                with open(path, 'a' if mode == "append" else 'w', encoding="utf-8") as f: f.write(new_content)
-                    f.write(new_content)
             yield f"[Status] ✅ {mode.capitalize()} 成功 ({len(new_content)} bytes)\n"
             next_prompt = self._get_anchor_prompt()
             return StepOutcome({"status": "success", 'writed_bytes': len(new_content)}, 

simphtml.py

@@ -1,7 +1,7 @@
 try: from bs4 import BeautifulSoup
 except ImportError: print("[Error] BeautifulSoup4 未安装，请叫Agent安装BeautifulSoup4，再使用web相关工具。")
 
-js_optHTML = r'''function optHTML() {
+js_optHTML = r'''function optHTML(text_only=false) {
 function createEnhancedDOMCopy() {  
   const nodeInfo = new WeakMap();  
   const ignoreTags = ['SCRIPT', 'STYLE', 'NOSCRIPT', 'META', 'LINK', 'COLGROUP', 'COL', 'TEMPLATE', 'PARAM', 'SOURCE'];  
@@ -43,6 +43,12 @@ function createEnhancedDOMCopy() {
         }
       } catch(e) {}
     }
+    if (sourceNode.shadowRoot) {
+      for (const shadowChild of sourceNode.shadowRoot.childNodes) {
+        const shadowClone = cloneNode(shadowChild, keep);
+        if (shadowClone) childNodes.push(shadowClone);
+      }
+    }
 
     const rect = sourceNode.getBoundingClientRect();
     const style = window.getComputedStyle(sourceNode);
@@ -90,7 +96,8 @@ function createEnhancedDOMCopy() {
     }  
   };  
 }  
-const { domCopy, getNodeInfo, isVisible } = createEnhancedDOMCopy();  
+const { domCopy, getNodeInfo, isVisible } = createEnhancedDOMCopy();
+if (text_only) return domCopy.innerText;  
 const viewportArea = window.innerWidth * window.innerHeight; 
 
 function analyzeNode(node, pPathType='main') {  
@@ -823,8 +830,8 @@ def get_temp_texts(driver):
         return []
     
 import time
-def get_main_block(driver, extra_js=""): 
+def get_main_block(driver, extra_js="", text_only=False): 
-    return driver.execute_js(extra_js+'\n'+js_optHTML).get('data', '')
+    return driver.execute_js(f"{extra_js}\n{js_optHTML}\nreturn optHTML({str(text_only).lower()});").get('data', '')
 
 def find_changed_elements(before_html, after_html):
     before_soup = BeautifulSoup(before_html, 'html.parser')
@@ -855,8 +862,9 @@ def find_changed_elements(before_html, after_html):
         result["top_change"] = h if len(h) <= 2000 else h[:2000] + '...[TRUNCATED]'
     return result
 
-def get_html(driver, cutlist=False, maxchars=38000, instruction="", extra_js=""):
+def get_html(driver, cutlist=False, maxchars=38000, instruction="", extra_js="", text_only=False):
-    page = get_main_block(driver, extra_js=extra_js)
+    page = get_main_block(driver, extra_js=extra_js, text_only=text_only)
+    if text_only: return page
     soup = optimize_html_for_tokens(page)
     html = str(soup)
     if not cutlist or len(html) <= maxchars: return html