refactor: agent loop done_hooks, inline_eval, scheduler logging & health_check

- agent_loop: for→while, _done_hooks callback mechanism, max_turns=40
- agentmain: pass self to handler, abort clears task queue
- stapp: heartbeat yield for Streamlit StopException detection
- ga: _inline_eval param for in-process eval, no_tool thresholds tuned
- scheduler: logging, max_delay_hours, weekday repeat, health_check()
This commit is contained in:
Liang Jiaqing
2026-03-30 13:51:21 +08:00
parent 7b704b5dbc
commit 473659ba6e
5 changed files with 131 additions and 37 deletions

View File

@@ -46,14 +46,15 @@ def get_pretty_json(data):
data["script"] = data["script"].replace("; ", ";\n ")
return json.dumps(data, indent=2, ensure_ascii=False).replace('\\n', '\n')
def agent_runner_loop(client, system_prompt, user_input, handler, tools_schema, max_turns=15, verbose=True, initial_user_content=None):
def agent_runner_loop(client, system_prompt, user_input, handler, tools_schema, max_turns=40, verbose=True, initial_user_content=None):
messages = [
{"role": "system", "content": system_prompt},
{"role": "user", "content": initial_user_content if initial_user_content is not None else user_input}
]
for turn in range(max_turns):
yield f"**LLM Running (Turn {turn+1}) ...**\n\n"
if (turn+1) % 10 == 0: client.last_tools = '' # 每10轮重置一次工具描述避免上下文过大导致的模型性能下降
turn = 0; handler._done_hooks = []; handler.max_turns = max_turns
while turn < handler.max_turns:
turn += 1; yield f"**LLM Running (Turn {turn}) ...**\n\n"
if turn%10 == 0: client.last_tools = '' # 每10轮重置一次工具描述避免上下文过大导致的模型性能下降
response_gen = client.chat(messages=messages, tools=tools_schema)
if verbose:
response = yield from response_gen
@@ -66,7 +67,7 @@ def agent_runner_loop(client, system_prompt, user_input, handler, tools_schema,
else: tool_calls = [{'tool_name': tc.function.name, 'args': json.loads(tc.function.arguments)}
for tc in response.tool_calls]
next_prompt = ""
next_prompt = ""; should_exit = None
for ii, tc in enumerate(tool_calls):
tool_name, args = tc['tool_name'], tc['args']
if tool_name == 'no_tool': pass
@@ -74,22 +75,25 @@ def agent_runner_loop(client, system_prompt, user_input, handler, tools_schema,
showarg = get_pretty_json(args)
if not verbose and len(showarg) > 200: showarg = showarg[:200] + ' ...'
yield f"🛠️ **正在调用工具:** `{tool_name}` 📥**参数:**\n````text\n{showarg}\n````\n"
handler.current_turn = turn + 1
handler.current_turn = turn
gen = handler.dispatch(tool_name, args, response, index=ii)
if verbose:
yield '`````\n'
outcome = yield from gen
yield '`````\n'
else: outcome = exhaust(gen)
if outcome.next_prompt is None: return {'result': 'CURRENT_TASK_DONE', 'data': outcome.data}
if outcome.should_exit: return {'result': 'EXITED', 'data': outcome.data}
if outcome.should_exit: return {'result': 'EXITED', 'data': outcome.data} # should_exit is only used for immediate exit
if not outcome.next_prompt:
should_exit = {'result': 'CURRENT_TASK_DONE', 'data': outcome.data}; break
if outcome.next_prompt.startswith('未知工具'): client.last_tools = ''
if outcome.data is not None:
datastr = json.dumps(outcome.data, ensure_ascii=False, default=json_default) if type(outcome.data) in [dict, list] else str(outcome.data)
next_prompt += f"<tool_result>\n{datastr}\n</tool_result>\n\n"
next_prompt += outcome.next_prompt
next_prompt = handler.next_prompt_patcher(next_prompt, None, turn+1)
next_prompt += outcome.next_prompt;
if not next_prompt:
if len(handler._done_hooks) == 0: return should_exit
next_prompt += handler._done_hooks.pop(0)
next_prompt = handler.next_prompt_patcher(next_prompt, None, turn)
messages = [{"role": "user", "content": next_prompt}] # just new message, history is kept in *Session
return {'result': 'MAX_TURNS_EXCEEDED'}