entari-plugin-hyw 3.3.8__py3-none-any.whl → 3.4.0__py3-none-any.whl

entari_plugin_hyw/core/pipeline.py

@@ -39,10 +39,8 @@ class ProcessingPipeline:
         self.client = AsyncOpenAI(base_url=self.config.base_url, api_key=self.config.api_key)
         self.all_web_results = [] # Cache for search results
         self.current_mode = "standard"  # standard | agent
-        # Independent ID counters for each type
-        self.search_id_counter = 0
-        self.page_id_counter = 0
-        self.image_id_counter = 0
+        # Global ID counter for all types (unified numbering)
+        self.global_id_counter = 0
 
         self.web_search_tool = {
             "type": "function",
@@ -628,7 +626,12 @@ class ProcessingPipeline:
             }
 
     def _parse_tagged_response(self, text: str) -> Dict[str, Any]:
-        """Parse response for references and page references reordered by appearance."""
+        """Parse response and auto-infer references from [N] citations in body text.
+        
+        New simplified format:
+        - Body text uses [1][2] format for citations
+        - No ref code block needed - we auto-infer from citations
+        """
         parsed = {"response": "", "references": [], "page_references": [], "image_references": [], "flow_steps": []}
         if not text:
             return parsed
@@ -646,104 +649,65 @@ class ProcessingPipeline:
         except Exception:
             pass
 
-        # 2. Extract references from text first (Order by appearance)
-        # Pattern matches [search:123], [page:123], [image:123]
-        pattern = re.compile(r'\[(search|page|image):(\d+)\]', re.IGNORECASE)
-        
-        matches = list(pattern.finditer(remaining_text))
+        # 2. Extract all [N] citations from body text (scan left to right for order)
+        body_pattern = re.compile(r'\[(\d+)\]')
+        id_order = []  # Preserve citation order
         
-        search_map = {}  # old_id_str -> new_id (int)
-        page_map = {}
-        image_map = {}
+        for match in body_pattern.finditer(remaining_text):
+            try:
+                id_val = int(match.group(1))
+                if id_val not in id_order:
+                    id_order.append(id_val)
+            except ValueError:
+                pass
+
+        # 3. Build references by looking up cited IDs in all_web_results
+        # Order by appearance in text
+        old_to_new_map = {}  # old_id -> new_id (for search & page only)
         
-        def process_ref(tag_type, old_id):
-            # Find in all_web_results
-            result_item = next((r for r in self.all_web_results if r.get("_id") == old_id and r.get("_type") == tag_type), None)
+        for old_id in id_order:
+            # Find in all_web_results by _id
+            result_item = next((r for r in self.all_web_results if r.get("_id") == old_id), None)
             
-            if not result_item:
-                return
+            if result_item:
+                entry = {
+                    "title": result_item.get("title", ""),
+                    "url": result_item.get("url", ""),
+                    "domain": result_item.get("domain", "")
+                }
                 
-            entry = {
-                "title": result_item.get("title", ""),
-                "url": result_item.get("url", ""),
-                "domain": result_item.get("domain", "")
-            }
-            if tag_type == "image":
-                 entry["thumbnail"] = result_item.get("thumbnail", "")
-
-            # Add to respective list and map
-            # Check maps to avoid duplicates
-            if tag_type == "search":
-                if str(old_id) not in search_map:
+                item_type = result_item.get("_type", "")
+                
+                # Auto-classify by type
+                if item_type == "search":
                     parsed["references"].append(entry)
-                    search_map[str(old_id)] = len(parsed["references"])
-            elif tag_type == "page":
-                if str(old_id) not in page_map:
+                    old_to_new_map[old_id] = len(parsed["references"])
+                elif item_type == "page":
                     parsed["page_references"].append(entry)
-                    page_map[str(old_id)] = len(parsed["page_references"])
-            elif tag_type == "image":
-                if str(old_id) not in image_map:
+                    old_to_new_map[old_id] = len(parsed["page_references"])
+                elif item_type == "image":
+                    # Collect image but don't add to map (will be stripped from text)
+                    entry["thumbnail"] = result_item.get("thumbnail", "")
                     parsed["image_references"].append(entry)
-                    image_map[str(old_id)] = len(parsed["image_references"])
+                    # Note: no old_to_new_map entry - image citations will be removed
 
-        # Pass 1: Text Body
-        for m in matches:
+        # 4. Replace [old_id] with [new_id] in text, or remove if image
+        def replace_id(match):
             try:
-                process_ref(m.group(1).lower(), int(m.group(2)))
-            except ValueError:
-                continue
-
-        # 3. Pass 2: References Block (Capture items missed in text)
-        ref_block_match = re.search(r'```references\s*(.*?)\s*```', remaining_text, re.DOTALL | re.IGNORECASE)
-        if ref_block_match:
-            ref_content = ref_block_match.group(1).strip()
-            remaining_text = remaining_text.replace(ref_block_match.group(0), "").strip()
-            
-            for line in ref_content.split("\n"):
-                line = line.strip()
-                if not line: continue
-                # Match [id] [type]
-                # e.g. [1] [image] ... or [image:1] ...
-                
-                # Check for [id] [type] format
-                id_match = re.match(r"^\[(\d+)\]\s*\[(search|page|image)\]", line, re.IGNORECASE)
-                if id_match:
-                    try:
-                         process_ref(id_match.group(2).lower(), int(id_match.group(1)))
-                    except ValueError:
-                        pass
+                old_id = int(match.group(1))
+                new_id = old_to_new_map.get(old_id)
+                if new_id is not None:
+                    return f"[{new_id}]"
                 else:
-                    # Check for [type:id] format in list
-                    alt_match = re.match(r"^\[(search|page|image):(\d+)\]", line, re.IGNORECASE)
-                    if alt_match:
-                        try:
-                            process_ref(alt_match.group(1).lower(), int(alt_match.group(2)))
-                        except ValueError:
-                            pass
-
-        # 4. Replace tags in text with new sequential IDs
-
-        # 4. Replace tags in text with new sequential IDs
-        def replace_tag(match):
-            tag_type = match.group(1).lower()
-            old_id = match.group(2)
-            
-            new_id = None
-            if tag_type == "search":
-                new_id = search_map.get(old_id)
-            elif tag_type == "page":
-                new_id = page_map.get(old_id)
-            elif tag_type == "image":
-                new_id = image_map.get(old_id)
-            
-            if new_id is not None:
-                if tag_type == "image":
-                    return ""
-                return f"[{tag_type}:{new_id}]"
-            
+                    # Check if it's an image reference (not in map)
+                    item = next((r for r in self.all_web_results if r.get("_id") == old_id), None)
+                    if item and item.get("_type") == "image":
+                        return ""  # Remove image citations from text
+            except ValueError:
+                pass
             return match.group(0)
 
-        remaining_text = pattern.sub(replace_tag, remaining_text)
+        remaining_text = body_pattern.sub(replace_id, remaining_text)
 
         parsed["response"] = remaining_text.strip()
         return parsed
@@ -766,10 +730,10 @@ class ProcessingPipeline:
             query = args.get("query")
             web = await self.search_service.search(query)
             
-            # Cache results and assign search-specific IDs
+            # Cache results and assign global IDs
             for item in web:
-                self.search_id_counter += 1
-                item["_id"] = self.search_id_counter
+                self.global_id_counter += 1
+                item["_id"] = self.global_id_counter
                 item["_type"] = "search"
                 item["query"] = query
                 self.all_web_results.append(item)
@@ -780,10 +744,10 @@ class ProcessingPipeline:
             query = args.get("query")
             images = await self.search_service.image_search(query)
 
-            # Cache results and assign image-specific IDs
+            # Cache results and assign global IDs
             for item in images:
-                self.image_id_counter += 1
-                item["_id"] = self.image_id_counter
+                self.global_id_counter += 1
+                item["_id"] = self.global_id_counter
                 item["_type"] = "image"
                 item["query"] = query
                 item["is_image"] = True
@@ -797,11 +761,11 @@ class ProcessingPipeline:
             # Returns Dict: {content, title, url}
             result_dict = await self.search_service.fetch_page(url)
             
-            # Cache the crawled content with page-specific ID
-            self.page_id_counter += 1
+            # Cache the crawled content with global ID
+            self.global_id_counter += 1
             
             cached_item = {
-                "_id": self.page_id_counter,
+                "_id": self.global_id_counter,
                 "_type": "page",
                 "title": result_dict.get("title", "Page"),
                 "url": result_dict.get("url", url),

entari_plugin_hyw/core/render.py

@@ -266,18 +266,27 @@ class ContentRenderer:
 
                 content_html = restore_math(content_html)
                 
-                # Convert [search:N] to blue badge
-                content_html = re.sub(
-                    r'\[search:(\d+)\]',
-                    r'<span class="inline-flex items-center justify-center min-w-[16px] h-4 px-0.5 text-[10px] font-bold text-blue-600 bg-blue-50 border border-blue-200 rounded mx-0.5 align-top relative -top-0.5">\1</span>',
-                    content_html
-                )
-                # Convert [page:N] to orange badge
-                content_html = re.sub(
-                    r'\[page:(\d+)\]',
-                    r'<span class="inline-flex items-center justify-center min-w-[16px] h-4 px-0.5 text-[10px] font-bold text-orange-700 bg-orange-50 border border-orange-200 rounded mx-0.5 align-top relative -top-0.5">\1</span>',
-                    content_html
-                )
+                # Convert [N] to colored badges based on index position
+                # - Numbers 1 to len(references) → blue (search results)
+                # - Numbers len(references)+1 to len(references)+len(page_references) → orange (page content)
+                
+                num_search_refs = len(references) if references else 0
+                num_page_refs = len(page_references) if page_references else 0
+                
+                def replace_badge(match):
+                    n = int(match.group(1))
+                    if 1 <= n <= num_search_refs:
+                        # Blue badge for search results
+                        return f'<span class="inline-flex items-center justify-center min-w-[16px] h-4 px-0.5 text-[10px] font-bold text-blue-600 bg-blue-50 border border-blue-200 rounded mx-0.5 align-top relative -top-0.5">{n}</span>'
+                    elif num_search_refs < n <= num_search_refs + num_page_refs:
+                        # Orange badge for page content (renumber from 1)
+                        page_num = n - num_search_refs
+                        return f'<span class="inline-flex items-center justify-center min-w-[16px] h-4 px-0.5 text-[10px] font-bold text-orange-700 bg-orange-50 border border-orange-200 rounded mx-0.5 align-top relative -top-0.5">{page_num}</span>'
+                    else:
+                        # Fallback: keep original if out of range
+                        return match.group(0)
+                
+                content_html = re.sub(r'\[(\d+)\]', replace_badge, content_html)
                 
                 # Strip out the references code block if it leaked into the content
                 content_html = re.sub(r'<pre><code[^>]*>.*?references.*?</code></pre>\s*$', '', content_html, flags=re.DOTALL | re.IGNORECASE)

entari_plugin_hyw/utils/prompts.py

@@ -40,7 +40,6 @@ Do not reply with extra text after calling tools to save tokens.
 ```
 """
 
-
 INSTRUCT_SP_VISION_ADD = """
 ## Vision Expert Message
 ```text
@@ -53,8 +52,6 @@ AGENT_SP = """# You are an Agent Control Expert. You need to understand user int
 
 Current Mode: {mode}, {mode_desc}
 
-
-
 ## Process Requirements
 When sending text without calling tools, it means this is the final reply. Please observe:
 - Provide a report directly, no need to explicitly answer the user message.
@@ -63,27 +60,11 @@ When sending text without calling tools, it means this is the final reply. Pleas
   - Use Markdown format, supporting highlight, katex.
   - Give a `# ` main title at the beginning, no extra nonsense, do not directly answer the user's question.
   - Rich content highlighting key points.
-- Tool Citation:
+- Citation:
   > Important: All body content must be based on actual information, ensuring 100% accuracy.
-  - Citation Rules:
-    - Cite information only if it is useful for solving the problem in this session; do not cite unnecessary messages.
-    - Badges must truly correspond to the information obtained in the context, and correspond to the content in references. Images correspond in order.
-  - Citation Rules in Body:
-    - Search Summary Citation: Use like `[search:3][search:4]`
-    - Page Content Citation: Use like `[page:5][page:6]`
-    - Image Citation: Use like `[image:7][image:8]`
-    - `search` means search summaries obtained using `internal_web_search`. Do not cite if no relevant info from this tool.
-    - `page` means page content obtained using `crawl_page`. Do not cite if no relevant info from this tool.
-    - `image` means images obtained using `internal_image_search`. Just place images in order, you do not need to explicitly cite.
-  - Add a `references` code block at the bottom of the body:
-    - Do not write unused entries. Do not write if experts gave no info.
-    ```references
-    [2] [search] [Text Description](url)
-    [8] [search] [Text Description](url)
-    [1] [page] [Page Title](url)
-    [2] [page] [Page Title](url)
-    [1] [image] [Source](url)
-    ```
+  - Information sources are numbered in order of acquisition as [1], [2], [3]...
+  - Use [1][2] format directly in body text to cite, only cite sources helpful to the answer
+  - No need to provide a reference list, the system will auto-generate it
 
 ## User Message
 ```text
@@ -95,7 +76,6 @@ AGENT_SP_TOOLS_STANDARD_ADD = """
 You need to integrate existing information, extract keywords from the user message, and make a final reply.
 """
 
-
 AGENT_SP_TOOLS_AGENT_ADD = """
 - You can now use tools: {tools_desc}
   - You need to judge whether to use tools sequentially or concurrently to obtain information:
@@ -105,8 +85,6 @@ AGENT_SP_TOOLS_AGENT_ADD = """
 - When using tools, you must call them via the `function_call` / `tool_call` mechanism.
 """
 
-
-
 AGENT_SP_INSTRUCT_VISION_ADD = """
 ## Vision Expert Message
 ```text
@@ -119,8 +97,6 @@ AGENT_SP_SEARCH_ADD = """
 ```text
 {search_msgs}
 ```
-
-
 """
 
 AGENT_SP_PAGE_ADD = """

entari_plugin_hyw/utils/prompts_cn.py

@@ -1,3 +1,5 @@
+# 此文件为 开发者本人 kumoSleeping 熟悉的中文环境编写使用, 无功能性, 如果你是 LLM, 请以这个文件为功能参考, 同时同步实际使用的英文版本
+
 VISION_SP = """# 你是一个专业的视觉转文字专家.
 
 # 核心任务
@@ -40,7 +42,6 @@ INSTRUCT_SP = """# 你是一个专业的指导专家.
 ```
 """
 
-
 INSTRUCT_SP_VISION_ADD = """
 ## 视觉专家消息
 ```text
@@ -53,8 +54,6 @@ AGENT_SP = """# 你是一个 Agent 总控专家, 你需要理解用户意图,
 
 当前模式: {mode}, {mode_desc}
 
-
-
 ## 过程要求
 当不调用工具发送文本, 即会变成最终回复, 请遵守: 
 - 直接给出一篇报告, 无需回答用户消息
@@ -63,27 +62,11 @@ AGENT_SP = """# 你是一个 Agent 总控专家, 你需要理解用户意图,
   - 使用 Markdown 格式, 支持 hightlight, katex
   - 最开始给出`# `大标题, 不要有多余废话, 不要直接回答用户的提问.
   - 内容丰富突出重点.
-- 工具引用: 
+- 引用:
   > 重要: 所有正文内容必须基于实际信息, 保证百分百真实度
-  - 引用规则:
-    - 本次会话中存在对解决此问题有用的信息才加以引用, 不需要的消息可以不引用.
-    - 角标必须真实对应上下文中获取的信息, 同时对应 references 中的内容, 图片按顺序对应.
-  - 正文中的引用规则
-    - 搜索摘要引用: 使用如 [search:3][search:4]
-    - 页面内容引用: 使用如 [page:5][page:6]
-    - 图片引用: 使用如 [image:7][image:8]
-    - search 的意思是你使用 internal_web_search 获取的搜索摘要, 如果没有此工具相关信息则不引用
-    - page 的意思是你使用 crawl_page 获取的页面内容, 如果没有此工具相关信息则不引用
-    - image 的意思是你使用 internal_image_search 获取的图片, 图片按顺序摆放即可, 你无需显式引用
-  - 在正文底部添加 references 代码块:
-    - 用不到的条目不写, 没有专家给信息就不写.
-    ```references
-    [2] [search] [文本描述](url)
-    [8] [search] [文本描述](url)
-    [1] [page] [页面标题](url)
-    [2] [page] [页面标题](url)
-    [1] [image] [来源](url)
-    ```
+  - 信息来源已按获取顺序编号为 [1], [2], [3]...
+  - 正文中直接使用 [1][2] 格式引用, 只引用对回答有帮助的来源
+  - 无需给出参考文献列表, 系统会自动生成
 
 ## 用户消息
 ```text
@@ -95,7 +78,6 @@ AGENT_SP_TOOLS_STANDARD_ADD = """
 你需要整合已有的信息, 提炼用户消息中的关键词, 进行最终回复.
 """
 
-
 AGENT_SP_TOOLS_AGENT_ADD = """
 - 你现在可以使用工具: {tools_desc}
   - 你需要判断顺序或并发使用工具获取信息:
@@ -105,8 +87,6 @@ AGENT_SP_TOOLS_AGENT_ADD = """
 - 使用工具时, 必须通过 function_call / tool_call 机制调用.
 """
 
-
-
 AGENT_SP_INSTRUCT_VISION_ADD = """
 ## 视觉专家消息
 ```text
@@ -119,8 +99,6 @@ AGENT_SP_SEARCH_ADD = """
 ```text
 {search_msgs}
 ```
-
-
 """
 
 AGENT_SP_PAGE_ADD = """

{entari_plugin_hyw-3.3.8.dist-info → entari_plugin_hyw-3.4.0.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: entari_plugin_hyw
-Version: 3.3.8
+Version: 3.4.0
 Summary: Use large language models to interpret chat messages
 Author-email: kumoSleeping <zjr2992@outlook.com>
 License: MIT

{entari_plugin_hyw-3.3.8.dist-info → entari_plugin_hyw-3.4.0.dist-info}/RECORD

@@ -33,16 +33,16 @@ entari_plugin_hyw/core/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3
 entari_plugin_hyw/core/config.py,sha256=uKZOuK9bG1W-E5UzhKX-EcYu2nFdxfl9EEaSwVgFtK4,1653
 entari_plugin_hyw/core/history.py,sha256=vqp7itwR5-KaqC4Ftmq6GOz7OM9GsiFJnSN9JJ2P6L4,5894
 entari_plugin_hyw/core/hyw.py,sha256=RCRjV9uYmvXysiliztphLP3VyUabrf0LY2Bk66W5JGA,1927
-entari_plugin_hyw/core/pipeline.py,sha256=6fA59ObZSV7Cb0mluOumpXve0m0WFGFgAopHVRgbah4,49197
-entari_plugin_hyw/core/render.py,sha256=rUhv2R5fdtsMIGg-Q1qe8hhUWC1_E50BODLA78u4_SI,28948
+entari_plugin_hyw/core/pipeline.py,sha256=5pU7K8q8jYMJVdudUtpz0Pq4bI1MUpBX0Jdrkc0vswE,47868
+entari_plugin_hyw/core/render.py,sha256=3tgmB3Pntbcr4YcyvF8tzaihNdol9sitFkXPrgmQXVQ,29696
 entari_plugin_hyw/utils/__init__.py,sha256=TnkxDqYr0zgRE7TC92tVbUaY8m1UyyoLg2zvzQ8nMVI,84
 entari_plugin_hyw/utils/browser.py,sha256=LJlFh-oSqt9mQBpMALxbYGUG__t1YLUo7RxUAslsWUc,1416
 entari_plugin_hyw/utils/misc.py,sha256=_7iHVYj_mJ6OGq6FU1s_cFeS1Ao-neBjZYd6eI2p95U,3482
 entari_plugin_hyw/utils/playwright_tool.py,sha256=ZZNkzFtUt_Gxny3Od4boBAgNF9J0N84uySatzn1Bwe4,1272
-entari_plugin_hyw/utils/prompts.py,sha256=eybCtSuW4F13jQtleHXF6CQypCIIutDT7mpbkXS48Gs,4993
-entari_plugin_hyw/utils/prompts_cn.py,sha256=87ti20ofjc8QW3i8HaPCUpDfl0EsS-ynz78e4tCz4Cg,4431
+entari_plugin_hyw/utils/prompts.py,sha256=Jp94gc0BE6Kn-5N2soXDo66ySX7NxrczCcuS16nGaMU,4035
+entari_plugin_hyw/utils/prompts_cn.py,sha256=DNSd4U6htUGcFXbywEX4d-WXnQL57CZsq-vba-T6hHw,3695
 entari_plugin_hyw/utils/search.py,sha256=Bvz2KFw3Gr2nuvmlo_8ExLHvO353NKX-YN35A2FCsBw,19047
-entari_plugin_hyw-3.3.8.dist-info/METADATA,sha256=a6rO4NRoNrbOG3MjRbFeCxCuYkm8OoIYpAaibPxszgM,3598
-entari_plugin_hyw-3.3.8.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-entari_plugin_hyw-3.3.8.dist-info/top_level.txt,sha256=TIDsn6XPs6KA5e3ezsE65JoXsy03ejDdrB41I4SPjmo,18
-entari_plugin_hyw-3.3.8.dist-info/RECORD,,
+entari_plugin_hyw-3.4.0.dist-info/METADATA,sha256=iLjguVLoNWNm3eMhE2cmGqDcZyW25wn3fUCHmU4idwE,3598
+entari_plugin_hyw-3.4.0.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+entari_plugin_hyw-3.4.0.dist-info/top_level.txt,sha256=TIDsn6XPs6KA5e3ezsE65JoXsy03ejDdrB41I4SPjmo,18
+entari_plugin_hyw-3.4.0.dist-info/RECORD,,