2026-09-13 16:38:32 +00:00
|
|
|
|
"""Opt-in paid-model team test. Existing credentials, disposable local workspace."""
|
|
|
|
|
|
import json,os,signal,socket,subprocess,tempfile,time
|
|
|
|
|
|
from pathlib import Path
|
|
|
|
|
|
BINARY=Path(__file__).resolve().parents[1]/'target/debug/grokboy'
|
|
|
|
|
|
def main():
|
|
|
|
|
|
with tempfile.TemporaryDirectory(prefix='gb-live-team-',dir='/tmp') as root:
|
|
|
|
|
|
root=Path(root);data=root/'data';workspace=root/'work';workspace.mkdir();(workspace/'items.txt').write_text('banana\napple\nbanana\npear\n')
|
|
|
|
|
|
env={**os.environ,'GROKBOY_DATA_DIR':str(data),'GROKBOY_MAX_ROUNDS_TOTAL':'16','GROKBOY_BROWSER_HEADED':'0'}
|
|
|
|
|
|
with (root/'daemon.log').open('w') as log:
|
|
|
|
|
|
p=subprocess.Popen([str(BINARY),'serve'],cwd=workspace,env=env,stdout=log,stderr=log)
|
|
|
|
|
|
def rpc(op,agent=None,**kw):
|
|
|
|
|
|
with socket.socket(socket.AF_UNIX) as s:
|
|
|
|
|
|
s.settimeout(10);s.connect(str(data/'service.sock'));s.sendall((json.dumps({'op':op,'agent':agent,**kw})+'\n').encode());v=json.loads(s.makefile().readline());assert 'error' not in v,v;return v
|
|
|
|
|
|
try:
|
|
|
|
|
|
for _ in range(100):
|
|
|
|
|
|
if (data/'service.sock').exists():break
|
|
|
|
|
|
assert p.poll() is None,(root/'daemon.log').read_text();time.sleep(.1)
|
|
|
|
|
|
for name in ('main','analyst'):rpc('create',name=name,cwd=str(workspace))
|
|
|
|
|
|
prompt='請用 delegate_task 把這份工作交給既有的 analyst agent，讓他在背景完成，你先簡短回覆已交辦即可：讀目前工作區 items.txt，去重並依字母排序寫成 clean.txt，每行一筆，再用 read_file 實際讀回確認。請讓 analyst 完成後回報產物路徑與驗證內容。不要上網，不要新增其他工作。'
|
|
|
|
|
|
rpc('chat','main',message=prompt);after=0;tasks=[];replied=False;pinged=False
|
|
|
|
|
|
deadline=time.monotonic()+300
|
|
|
|
|
|
while time.monotonic()<deadline:
|
|
|
|
|
|
for e in rpc('events','main',after=after)['events']:
|
|
|
|
|
|
after=e['id'];kind=e['kind'];payload=e['payload']
|
|
|
|
|
|
if kind in ('reply','task_queued','task_ended'):print(kind,json.dumps(payload,ensure_ascii=False),flush=True)
|
|
|
|
|
|
if kind=='reply' and tasks and all(t['state']=='terminal' for t in tasks):replied=True
|
|
|
|
|
|
tasks=rpc('tasks','main')
|
|
|
|
|
|
if tasks and not pinged:rpc('chat','main',message='謝謝，我還可以在這裡繼續聊天吧？不用改動背景工作。');pinged=True
|
|
|
|
|
|
roots=[t for t in tasks if t['parent_id'] is None]
|
|
|
|
|
|
if roots and all(t['state']=='terminal' for t in roots):
|
|
|
|
|
|
if not all(t['verdict'] in ('done','answer') for t in roots):
|
|
|
|
|
|
import sqlite3
|
|
|
|
|
|
db=sqlite3.connect(data/'team.sqlite3')
|
|
|
|
|
|
for (raw,) in db.execute('SELECT data FROM tasks'):
|
|
|
|
|
|
saved=json.loads(raw)
|
|
|
|
|
|
for m in saved['session']['messages']:
|
|
|
|
|
|
if m.get('tool_calls'):print('calls:',[c['function']['name'] for c in m['tool_calls']],flush=True)
|
|
|
|
|
|
if m['role']=='tool' and 'error' in m.get('content',''):print('tool feedback:',m['content'],flush=True)
|
|
|
|
|
|
raise AssertionError(roots)
|
|
|
|
|
|
if replied:break
|
|
|
|
|
|
time.sleep(.5)
|
|
|
|
|
|
else:raise AssertionError(('timed out',tasks))
|
|
|
|
|
|
assert (workspace/'clean.txt').read_text().splitlines()==['apple','banana','pear']
|
|
|
|
|
|
assert tasks and all(t['requests']<=t['limit'] for t in tasks)
|
|
|
|
|
|
import sqlite3
|
|
|
|
|
|
db=sqlite3.connect(data/'team.sqlite3');saved=[json.loads(row[0]) for row in db.execute('SELECT data FROM tasks')]
|
2026-09-14 09:08:35 +00:00
|
|
|
|
assert any(any(call['function']['name']=='external_read_file' and 'clean.txt' in call['function']['arguments'] for message in t['session']['messages'] for call in message.get('tool_calls',[])) for t in saved)
|
2026-09-13 16:38:32 +00:00
|
|
|
|
print('PASS live model: existing-agent delegation, concurrent chat, verified artifact and returned report',flush=True)
|
|
|
|
|
|
print('Root task model requests:',[t['requests'] for t in tasks if t['parent_id'] is None],flush=True)
|
|
|
|
|
|
finally:
|
|
|
|
|
|
if p.poll() is None:p.send_signal(signal.SIGINT);p.wait(timeout=10)
|
|
|
|
|
|
if __name__=='__main__':main()
|
