feat: 实现推理过程实时可视化（WebSocket事件推送）
- 新增 per-user WebSocket 回调机制，支持多用户并发推理会话 - 在每个推理阶段推送实时事件：context/plan/safety/execution/tool_call - reasoning_console.wss 使用 ws_push_callbacks 字典替代共享 ws_push - 新增 _current_user_id/_current_org_id 追踪用于用户隔离 - _execute_tool 传递 context 参数确保工具执行的用户隔离 - 新增文件技能搜索：扫描用户目录和共享目录的 skills，与DB技能去重
This commit is contained in:
parent
766bd9ecb9
commit
89e099ee12
@ -101,8 +101,8 @@ TOOL_DESCRIPTIONS = "\n".join(f"- {t['name']}: {t['desc']}" for t in AVAILABLE_T
|
||||
class HermesReasoningEngine:
|
||||
"""Production reasoning engine that uses LLM and real tool execution."""
|
||||
|
||||
# Websocket push callback (injected by .wss endpoint)
|
||||
ws_push = None
|
||||
# Websocket push callbacks keyed by user_id (injected by .wss endpoint)
|
||||
ws_push_callbacks: Dict[str, callable] = {}
|
||||
|
||||
DEFAULT_SAFETY_RULES = {
|
||||
"strict": [
|
||||
@ -116,20 +116,22 @@ class HermesReasoningEngine:
|
||||
}
|
||||
|
||||
def __init__(self):
|
||||
pass
|
||||
self._current_user_id = None # Set during reason_and_execute execution
|
||||
self._current_org_id = None # Set during reason_and_execute execution
|
||||
|
||||
async def _push(self, event_type: str, data: Dict[str, Any] = None):
|
||||
"""Push a reasoning step event via websocket."""
|
||||
if self.ws_push:
|
||||
async def _push(self, event_type: str, data: Dict[str, Any] = None, user_id: str = None):
|
||||
"""Push a reasoning step event via websocket for a specific user."""
|
||||
if user_id and user_id in self.ws_push_callbacks:
|
||||
ws_push = self.ws_push_callbacks[user_id]
|
||||
msg = {
|
||||
'event': event_type,
|
||||
'data': data or {},
|
||||
'timestamp': time.time(),
|
||||
}
|
||||
try:
|
||||
await self.ws_push(msg)
|
||||
await ws_push(msg)
|
||||
except Exception as e:
|
||||
error(f"ws_push failed: {e}")
|
||||
error(f"ws_push failed for user {user_id}: {e}")
|
||||
|
||||
# --------------------------------------------------------
|
||||
# Config helpers
|
||||
@ -178,14 +180,14 @@ class HermesReasoningEngine:
|
||||
|
||||
async def _get_memory_context(self, user_id: str, request: str, config: Dict) -> Dict[str, Any]:
|
||||
"""Get real memory and session context from harnessed_agent."""
|
||||
context = {"user_id": user_id, "memory_entries": [], "recent_sessions": [], "skills": []}
|
||||
context = {"user_id": user_id, "org_id": self._current_org_id, "memory_entries": [], "recent_sessions": [], "skills": []}
|
||||
|
||||
await self._push('context_start', {'message': '正在收集上下文...', 'user_id': user_id})
|
||||
await self._push('context_start', {'message': '正在收集上下文...', 'user_id': user_id}, user_id=self._current_user_id)
|
||||
|
||||
try:
|
||||
# Intelligent memory
|
||||
max_tokens = int(config.get('max_context_tokens', 4000)) // 3
|
||||
await self._push('context_memory', {'message': '加载记忆上下文...', 'max_tokens': max_tokens})
|
||||
await self._push('context_memory', {'message': '加载记忆上下文...', 'max_tokens': max_tokens}, user_id=self._current_user_id)
|
||||
if hasattr(ServerEnv(), 'harnessed_get_intelligent_memory_context'):
|
||||
mem_result = await ServerEnv().harnessed_get_intelligent_memory_context(
|
||||
current_task=request,
|
||||
@ -196,7 +198,7 @@ class HermesReasoningEngine:
|
||||
await self._push('context_memory_done', {
|
||||
'message': f'加载 {len(context["memory_entries"])} 条记忆',
|
||||
'count': len(context['memory_entries'])
|
||||
})
|
||||
}, user_id=self._current_user_id)
|
||||
|
||||
# Session search
|
||||
if config.get('enable_cross_session_search', '1') == '1':
|
||||
@ -217,14 +219,50 @@ class HermesReasoningEngine:
|
||||
|
||||
return context
|
||||
|
||||
def _search_skill_dir(self, skills_dir: str, keywords: set, source: str = "user") -> List[Dict[str, Any]]:
|
||||
"""Search a skill directory for skills matching keywords."""
|
||||
import os
|
||||
results = []
|
||||
if not os.path.exists(skills_dir):
|
||||
return results
|
||||
for skill_name in os.listdir(skills_dir):
|
||||
skill_path = os.path.join(skills_dir, skill_name)
|
||||
skill_file = os.path.join(skill_path, "SKILL.md")
|
||||
if os.path.isdir(skill_path) and os.path.exists(skill_file):
|
||||
matches = False
|
||||
for kw in keywords:
|
||||
if kw in skill_name.lower():
|
||||
matches = True
|
||||
break
|
||||
if not matches:
|
||||
with open(skill_file, 'r', encoding='utf-8', errors='ignore') as f:
|
||||
content = f.read().lower()
|
||||
for kw in keywords:
|
||||
if kw in content:
|
||||
matches = True
|
||||
break
|
||||
if matches:
|
||||
results.append({
|
||||
'id': f'{source}_{skill_name}',
|
||||
'name': skill_name,
|
||||
'description': f'{source} skill: {skill_name}',
|
||||
'content_preview': '',
|
||||
'source': source
|
||||
})
|
||||
return results
|
||||
|
||||
async def _find_relevant_skills(self, user_id: str, request: str) -> List[Dict[str, Any]]:
|
||||
"""Find skills relevant to the request via keyword search."""
|
||||
"""Find skills relevant to the request via keyword search.
|
||||
Searches both DB-based skills and file-based user skills directory."""
|
||||
import os
|
||||
keywords = set()
|
||||
for word in request.lower().split():
|
||||
if len(word) > 2:
|
||||
keywords.add(word)
|
||||
|
||||
skills = []
|
||||
|
||||
# 1. Search DB-based skills (harnessed_agent module's manage_skills)
|
||||
try:
|
||||
env = ServerEnv()
|
||||
dbname = env.get_module_dbname('harnessed_reasoning')
|
||||
@ -242,17 +280,34 @@ class HermesReasoningEngine:
|
||||
})
|
||||
rows = (rows or [])[:2]
|
||||
skills.extend(rows)
|
||||
|
||||
# Deduplicate
|
||||
seen = set()
|
||||
unique = []
|
||||
for s in skills:
|
||||
if s['id'] not in seen:
|
||||
seen.add(s['id'])
|
||||
unique.append(s)
|
||||
return unique[:5]
|
||||
except Exception:
|
||||
return []
|
||||
pass
|
||||
|
||||
# 2. Search file-based user skills directory (~/.hermes/users/{user_id}/skills/)
|
||||
try:
|
||||
hermes_dir = os.path.expanduser("~/.hermes")
|
||||
user_skills_dir = os.path.join(hermes_dir, "users", str(user_id), "skills")
|
||||
skills.extend(self._search_skill_dir(user_skills_dir, keywords, source="user"))
|
||||
except Exception:
|
||||
pass
|
||||
|
||||
# 3. Search shared skills directory (~/.hermes/skills/)
|
||||
try:
|
||||
hermes_dir = os.path.expanduser("~/.hermes")
|
||||
shared_skills_dir = os.path.join(hermes_dir, "skills")
|
||||
skills.extend(self._search_skill_dir(shared_skills_dir, keywords, source="shared"))
|
||||
except Exception:
|
||||
pass
|
||||
|
||||
# Deduplicate by name
|
||||
seen = set()
|
||||
unique = []
|
||||
for s in skills:
|
||||
name = s.get('name', '')
|
||||
if name not in seen:
|
||||
seen.add(name)
|
||||
unique.append(s)
|
||||
return unique[:5]
|
||||
|
||||
# --------------------------------------------------------
|
||||
# Safety checks
|
||||
@ -315,7 +370,7 @@ class HermesReasoningEngine:
|
||||
async def _generate_plan(self, request: str, context: Dict[str, Any],
|
||||
config: Dict[str, Any]) -> Dict[str, Any]:
|
||||
"""Use LLM to analyze request and generate execution plan."""
|
||||
await self._push('plan_start', {'message': 'LLM 正在分析请求并生成执行计划...', 'request': request[:100]})
|
||||
await self._push('plan_start', {'message': 'LLM 正在分析请求并生成执行计划...', 'request': request[:100]}, user_id=self._current_user_id)
|
||||
|
||||
# Build context summary
|
||||
ctx_parts = []
|
||||
@ -352,7 +407,7 @@ class HermesReasoningEngine:
|
||||
|
||||
if 'error' in result:
|
||||
error(f"LLM planning failed: {result['error'].get('message')}")
|
||||
await self._push('plan_error', {'message': f'LLM 调用失败: {result["error"].get("message")}'})
|
||||
await self._push('plan_error', {'message': f'LLM 调用失败: {result["error"].get("message")}'}, user_id=self._current_user_id)
|
||||
return {
|
||||
'analysis': 'LLM 调用失败，无法生成计划',
|
||||
'steps': [],
|
||||
@ -370,7 +425,7 @@ class HermesReasoningEngine:
|
||||
'analysis': plan.get('analysis', ''),
|
||||
'step_count': steps_count,
|
||||
'steps': plan.get('steps', [])
|
||||
})
|
||||
}, user_id=self._current_user_id)
|
||||
return plan
|
||||
|
||||
def _parse_plan_json(self, text: str) -> Dict[str, Any]:
|
||||
@ -439,6 +494,7 @@ class HermesReasoningEngine:
|
||||
return await env.harnessed_execute_tool(
|
||||
tool_name=tool_name,
|
||||
parameters=parameters,
|
||||
context=context,
|
||||
)
|
||||
|
||||
# Fallback: try to call the tool function directly if registered on ServerEnv
|
||||
@ -473,6 +529,16 @@ class HermesReasoningEngine:
|
||||
if not user_id:
|
||||
user_id = "anonymous"
|
||||
|
||||
self._current_user_id = user_id # Set for user-isolated ws_push
|
||||
|
||||
# Get org_id from ServerEnv for shared skill permission checks
|
||||
self._current_org_id = None
|
||||
try:
|
||||
env = ServerEnv()
|
||||
self._current_org_id = getattr(env, 'orgid', None) or getattr(env, 'org_id', None)
|
||||
except Exception:
|
||||
pass
|
||||
|
||||
config = await self._get_config(user_id)
|
||||
safety_mode = config.get('safety_mode', 'strict')
|
||||
|
||||
@ -492,14 +558,14 @@ class HermesReasoningEngine:
|
||||
'user_id': user_id,
|
||||
'request': request,
|
||||
'message': '推理引擎启动'
|
||||
})
|
||||
}, user_id=self._current_user_id)
|
||||
context = await self._get_memory_context(user_id, request, config)
|
||||
context['user_id'] = user_id # Ensure user_id is available for tool execution
|
||||
|
||||
await self._push('context_complete', {
|
||||
'message': self._context_summary(context),
|
||||
'summary': self._context_summary(context)
|
||||
})
|
||||
}, user_id=self._current_user_id)
|
||||
|
||||
# Step 2: LLM-based planning
|
||||
plan = await self._generate_plan(request, context, config)
|
||||
@ -514,9 +580,9 @@ class HermesReasoningEngine:
|
||||
await self._push('safety_violation', {
|
||||
'violations': violations,
|
||||
'message': f'安全检查发现 {len(violations)} 个违规'
|
||||
})
|
||||
}, user_id=self._current_user_id)
|
||||
else:
|
||||
await self._push('safety_pass', {'message': '安全检查通过'})
|
||||
await self._push('safety_pass', {'message': '安全检查通过'}, user_id=self._current_user_id)
|
||||
|
||||
# Step 4: Store session
|
||||
await self._store_session(session_id, user_id, request, plan, violations, "planned")
|
||||
@ -565,12 +631,15 @@ class HermesReasoningEngine:
|
||||
await self._push('reasoning_error', {
|
||||
'error': str(e),
|
||||
'message': f'推理失败: {str(e)}'
|
||||
})
|
||||
}, user_id=user_id)
|
||||
|
||||
try:
|
||||
await self._update_session_status(session_id, "failed")
|
||||
except Exception:
|
||||
pass
|
||||
finally:
|
||||
self._current_user_id = None # Clean up
|
||||
self._current_org_id = None # Clean up
|
||||
|
||||
return result
|
||||
|
||||
@ -610,7 +679,7 @@ class HermesReasoningEngine:
|
||||
await self._push('execution_start', {
|
||||
'message': f'开始执行计划，共 {len(steps)} 个步骤',
|
||||
'total_steps': len(steps)
|
||||
})
|
||||
}, user_id=self._current_user_id)
|
||||
|
||||
for step in steps[:max_steps]:
|
||||
step_num = step.get('step_number', '?')
|
||||
@ -621,7 +690,7 @@ class HermesReasoningEngine:
|
||||
'step_number': step_num,
|
||||
'description': step_desc,
|
||||
'message': f'步骤 {step_num}: {step_desc}'
|
||||
})
|
||||
}, user_id=self._current_user_id)
|
||||
|
||||
for action in step.get('actions', [])[:max_tools]:
|
||||
tool = action.get('tool', '')
|
||||
@ -635,7 +704,7 @@ class HermesReasoningEngine:
|
||||
'tool': tool,
|
||||
'parameters': params,
|
||||
'message': f'调用工具: {tool}'
|
||||
})
|
||||
}, user_id=self._current_user_id)
|
||||
|
||||
info(f"Executing step {step_num}: {tool}({json.dumps(params, ensure_ascii=False)[:100]})")
|
||||
tool_result = await self._execute_tool(tool, params, context)
|
||||
@ -646,7 +715,7 @@ class HermesReasoningEngine:
|
||||
'success': tool_result.get('success', False),
|
||||
'result': str(tool_result)[:1000],
|
||||
'message': f'工具 {tool} 执行{"成功" if tool_result.get("success") else "失败"}'
|
||||
})
|
||||
}, user_id=self._current_user_id)
|
||||
|
||||
step_results.append({
|
||||
'tool': tool,
|
||||
@ -674,12 +743,12 @@ class HermesReasoningEngine:
|
||||
'description': step.get('description', ''),
|
||||
'action_count': len(step_results),
|
||||
'message': f'步骤 {step_num} 完成，执行了 {len(step_results)} 个操作'
|
||||
})
|
||||
}, user_id=self._current_user_id)
|
||||
|
||||
await self._push('execution_complete', {
|
||||
'message': f'计划执行完成，共 {len(all_results)} 个步骤',
|
||||
'total_steps': len(all_results)
|
||||
})
|
||||
}, user_id=self._current_user_id)
|
||||
|
||||
return all_results
|
||||
|
||||
|
||||
@ -89,11 +89,8 @@ async def _run_reasoning(user_id, request_text):
|
||||
|
||||
engine = get_harnessed_reasoning_engine()
|
||||
|
||||
# 注入 ws_push 回调到引擎实例
|
||||
async def push_callback(msg):
|
||||
await _ws_push(user_id, msg)
|
||||
|
||||
engine.ws_push = push_callback
|
||||
# 注册 per-user ws_push 回调到引擎
|
||||
engine.ws_push_callbacks[user_id] = lambda msg: _ws_push(user_id, msg)
|
||||
|
||||
try:
|
||||
result = await engine.reason_and_execute(
|
||||
@ -122,5 +119,5 @@ async def _run_reasoning(user_id, request_text):
|
||||
}
|
||||
})
|
||||
finally:
|
||||
# 清理回调
|
||||
engine.ws_push = None
|
||||
# 清理 per-user 回调
|
||||
engine.ws_push_callbacks.pop(user_id, None)
|
||||
|
||||
Loading…
x
Reference in New Issue
Block a user