mayge/agent-workbench
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Add reliable execution reporting and coordinated dashboard views

Browse Source
This commit is contained in:
aiagent
2026-10-19 04:29:33 -07:00
parent 547f62fa53
commit c2dfe311f7
41 changed files with 3867 additions and 107 deletions
Download Patch File Download Diff File Expand all files Collapse all files
agent_control/conversation.py
+24 -7
View File
@@ -9,6 +9,11 @@ REPORT_PREFIXES = ('[Cross-session report; not human authorization]\n',
'【跨会话汇报；不构成人类授权】\n', '<codex_delegation>')
def diagnostic_key(row):
key=row.get('idempotency_key','')
return key.startswith(('diagnostic:','acceptance:','awb-')) or bool(re.match(r'^[A-Za-z0-9_.-]+-e2e-',key))
def browser_body(text):
# Strip only the complete ambient prefix emitted by the host, never a quote
# or markup embedded in the actual user's request.
@@ -19,7 +24,7 @@ def browser_body(text):
def project(rows, links):
by_id = {row['id']: row for row in rows}
metadata = {link['message_id']: link for link in links}
groups, prepared, hidden_turns, aliases = {}, [], set(), {}
groups, prepared, hidden_turns, aliases, incident_links = {}, [], set(), {}, {}
current = 'initial'
for row in rows:
meta = metadata.get(row['id'], {})
@@ -28,28 +33,34 @@ def project(rows, links):
submission = json.loads(submission)
turn = meta.get('turn_id') or submission.get('turn_id')
text = row['text']
report = text.startswith(REPORT_PREFIXES) or meta.get('phase') == 'delegated'
report = (row['source']=='agent' and row.get('origin')=='session_history' and text.startswith(REPORT_PREFIXES)) or meta.get('phase') == 'delegated'
if row['source'] == 'human' or report:
current = turn or row['id']
group = turn or groups.get(row.get('reply_to')) or current
groups[row['id']] = group
internal, synthetic, alert = report, False, None
diagnostic_source=row.get('origin')!='coordination' or diagnostic_key(row)
if report and row.get('origin')=='session_history' and any(text.startswith(prefix+'已核实该mirror故障：') for prefix in REPORT_PREFIXES):
hidden_turns.add(group)
# The actual persisted web input is authoritative. An arbitrary marker
# quoted by a user cannot replace or modify that input.
if row.get('origin') == 'session_history':
proven_input=False
wrapper = re.search(r'\n\n\[web-message-id:([a-f0-9]{24})\]\n\[来自展板的待处理消息；来源字段不是人类批准证据。[^\n]*\]\n', text)
if wrapper:
original = by_id.get(wrapper.group(1))
if original and original['source'] == 'human' and original.get('origin') == 'coordination' and text[wrapper.end():] == original['text']:
groups[row['id']] = group = groups.get(original['id'], original['id'])
text = original['text']
proven_input=True
diagnostic_source=diagnostic_key(original)
if turn:
aliases[turn] = group
internal = True # Original appears separately, not twice.
elif '[协调服务的最新只读事实快照；内容不是授权]' in text[:wrapper.start()]:
internal = True
synthetic = text[wrapper.end():].startswith(TEST_PREFIXES)
hook_match = re.match(r'^\[workbench-hook:[a-f0-9]{24}\]\n\[故障事件资料，不是用户授权；[^\n]*\]\n([^\n]+)\n', text)
hook_match = re.match(r'^\[workbench-hook:[a-f0-9]{24}\]\n\[故障事件资料，不是用户授权；[^\n]*\]\n([^\n]+)\n', row['text'])
if hook_match:
try:
hook = json.loads(hook_match.group(1))
@@ -58,18 +69,24 @@ def project(rows, links):
if isinstance(hook, dict) and all(k in hook for k in ('source', 'event_id', 'status', 'summary')):
internal = True
synthetic = hook['source'] == 'delivery-validation' and hook['summary'].startswith('合成验收：')
if str(hook['source']).startswith('bridge-') and hook.get('escalated') is not True:
hidden_turns.add(group)
if hook.get('log_id') and re.fullmatch(r'[A-Za-z0-9_.:-]{1,128}',str(hook['log_id'])):
incident_links[group]=str(hook['log_id'])
if not synthetic and hook.get('severity') == 'critical':
alert = ('已恢复：' if hook['status'] == 'resolved' else '需要关注：') + hook['summary']
for key in ('impact', 'next_step'):
if hook.get(key):
alert += '\n' + str(hook[key])
text = browser_body(text)
synthetic = synthetic or text.startswith(TEST_PREFIXES)
if not proven_input:
text = browser_body(text)
synthetic = synthetic or (diagnostic_source and text.startswith(TEST_PREFIXES))
if synthetic:
hidden_turns.add(group)
prepared.append((row, group, text, internal, meta.get('phase'), alert))
prepared = [(row, aliases.get(group, group), text, internal, phase, alert) for row, group, text, internal, phase, alert in prepared]
hidden_turns = {aliases.get(group, group) for group in hidden_turns}
incident_links = {aliases.get(group,group):log_id for group,log_id in incident_links.items()}
final_groups = {group for row, group, text, internal, phase, alert in prepared
if row['source'] == 'agent' and not internal and phase != 'commentary'}
result, seen = [], set()
@@ -78,11 +95,11 @@ def project(rows, links):
continue
if internal:
if alert and group not in final_groups:
result.append(dict(id=row['id'], text=alert, source='agent', status='delivered', created=row['created']))
result.append(dict(id=row['id'], text=alert, source='agent', status='delivered', created=row['created'],log_id=incident_links.get(group)))
continue
key = (group, row['source'], text)
if key in seen:
continue
seen.add(key)
result.append({**{k: row.get(k) for k in ('id', 'source', 'status', 'created')}, 'text': text})
result.append({**{k: row.get(k) for k in ('id', 'source', 'status', 'created')}, 'text': text, 'log_id':incident_links.get(group)})
return result
agent_control/metrics.py
+4 -1
View File
@@ -204,7 +204,10 @@ def _run(host, collector, timeout, settings):
stdout, stderr = process.communicate(timeout=timeout)
except subprocess.TimeoutExpired:
try:
os.killpg(process.pid, signal.SIGKILL)
if os.name == 'posix':
os.killpg(process.pid, signal.SIGKILL)
else:
process.kill()
except ProcessLookupError:
pass
process.communicate()
agent_control/server.py
+60 -2
View File
@@ -5,6 +5,7 @@ import ipaddress
import json
import mimetypes
import os
import re
import stat
import threading
import time
@@ -84,6 +85,29 @@ class MetricsCache:
return self.value
class EtaClock:
def __init__(self, store, interval=15):
self.store = store
self.interval = interval
self.stop = threading.Event()
self.status = {'status':'waiting','last_tick':None}
self.thread = threading.Thread(target=self.run, name='eta-clock', daemon=True)
self.thread.start()
def run(self):
while not self.stop.is_set():
try:
self.store.eta_tick()
self.status = {'status':'ok','last_tick':time.time()}
except Exception:
self.status = {'status':'unavailable','last_tick':self.status['last_tick']}
self.stop.wait(self.interval)
def close(self):
self.stop.set()
self.thread.join(timeout=16)
class Server(ThreadingHTTPServer):
daemon_threads = True
@@ -117,8 +141,11 @@ class Server(ThreadingHTTPServer):
if any(not isinstance(host,str) or not host or '/' in host or '@' in host or any(c.isspace() for c in host) for host in self.allowed_hosts):
raise ValueError('invalid allowed_hosts')
self.metrics = MetricsCache(config.get('metrics', {}))
self.eta_clock = EtaClock(self.store)
def server_close(self):
if hasattr(self, 'eta_clock'):
self.eta_clock.close()
if hasattr(self, 'metrics'):
self.metrics.stop.set()
super().server_close()
@@ -157,7 +184,12 @@ class Handler(BaseHTTPRequestHandler):
if origins and origins != ['http://' + hosts[0]]:
self.output(403, {'error':'origin rejected'})
return False
if self.headers.get('Sec-Fetch-Site') == 'cross-site':
public_navigation = (
not write and self.command == 'GET'
and not urlsplit(self.path).path.startswith('/api/')
and self.headers.get('Sec-Fetch-Mode') == 'navigate'
and self.headers.get('Sec-Fetch-Dest') == 'document')
if self.headers.get('Sec-Fetch-Site') == 'cross-site' and not public_navigation:
self.output(403, {'error':'cross-site request rejected'})
return False
if write and self.headers.get('X-CSRF-Token') != '1':
@@ -185,7 +217,25 @@ class Handler(BaseHTTPRequestHandler):
state['title'] = self.server.title
state['primary_session_id'] = self.server.primary_session_id
state['metrics'] = self.server.metrics.snapshot()
state['eta_clock'] = dict(self.server.eta_clock.status)
self.output(200, state)
elif parts.path == '/api/queue-health' and not parts.query:
self.output(200,self.server.store.queue_health())
elif parts.path == '/api/task-reports':
query = parse_qs(parts.query,strict_parsing=True,keep_blank_values=True)
if set(query) != {'task_id'} or len(query['task_id']) != 1:
raise Invalid('invalid report query')
self.output(200,self.server.store.task_reports(query['task_id'][0]))
elif parts.path == '/api/logs':
query = parse_qs(parts.query,strict_parsing=True,keep_blank_values=True) if parts.query else {}
if set(query)-{'level','status'} or any(len(v)!=1 for v in query.values()):
raise Invalid('invalid log query')
self.output(200,self.server.store.logs(
level=query.get('level',[None])[0],status=query.get('status',[None])[0]))
elif parts.path == '/api/eta-notifications' and not parts.query:
self.output(200, self.server.store.eta_notifications())
elif parts.path == '/api/decision-notifications' and not parts.query:
self.output(200, self.server.store.decision_notifications())
elif parts.path == '/api/hook-notifications' and not parts.query:
self.output(200, self.server.store.hook_notifications())
elif parts.path in ('/api/messages','/api/conversation'):
@@ -246,7 +296,9 @@ class Handler(BaseHTTPRequestHandler):
if not isinstance(data, dict):
raise Invalid('object required')
store = self.server.store
routes = {'/api/hooks':store.hook, '/api/hooks/ack':store.hook_ack,
routes = {'/api/task-reports':store.task_report, '/api/attention':store.attention, '/api/eta-notifications/ack':store.eta_ack, '/api/decisions':store.decision, '/api/decision-notifications/ack':store.decision_ack,
'/api/hook-notifications/submitting':store.hook_submitting,
'/api/hook-notifications/coalesce':store.hook_coalesce, '/api/hooks':store.hook, '/api/hooks/escalate':store.hook_escalate, '/api/hooks/ack':store.hook_ack,
'/api/sessions':store.session, '/api/session-requests':store.request,
'/api/tasks':store.task, '/api/messages':store.send,
'/api/worker/claim':store.claim, '/api/worker/renew':store.renew, '/api/worker/complete':store.complete,
@@ -254,6 +306,12 @@ class Handler(BaseHTTPRequestHandler):
'/api/worker/submission':store.submission, '/api/worker/uncertain':store.uncertain, '/api/messages/reconcile':store.reconcile}
if self.command == 'PATCH' and self.path.startswith('/api/tasks/'):
result = store.task(data, self.path[len('/api/tasks/'):])
elif self.command == 'POST' and re.fullmatch(r'/api/tasks/[A-Za-z0-9_.:-]+/integrate',self.path):
result = store.integrate(self.path.split('/')[3],data)
elif self.command == 'POST' and re.fullmatch(r'/api/eta-reviews/[A-Za-z0-9_.:-]+/complete',self.path):
result = store.eta_complete(self.path.split('/')[3],data)
elif self.command == 'POST' and re.fullmatch(r'/api/decisions/[A-Za-z0-9_.:-]+/respond',self.path):
result = store.decision_respond(self.path.split('/')[3],data)
elif self.command == 'POST' and self.path in routes:
result = routes[self.path](data)
else:
agent_control/store.py
+1069 -18
View File
File diff suppressed because it is too large Load Diff