Files
aiturk-hermes-ide/apps/desktop/assets/codex-bridge/test_karar_asamalari.py
T
yilsemandClaude Opus 5.5 7c4ef18df0 feat(codex-bridge): decision-model stages, Dify rules and beta.39
Owner request, 14 October 2026: use the decision model (Jev / Celeris) at
several stages, let Dify supply the rules, code it without new bugs.
Every new stage is switched on in Settings > Kademeli Gecis and starts
disabled, so the upgrade changes no behaviour by itself.

- Settings schema 2 (Python and TypeScript validate the same rules); a
  schema 1 file is filled from the defaults. Default chain is now
  Katman-1 -> Katman-2 -> Katman-3 with Katman-3, Chat-Model and
  bekci-sesli-bekci always passing (owner decision of the same day).
- watchModels: the Bekci correction channel follows a settings list
  instead of the hard-coded Sesli-Hizli-Zeki.
- prefilter: a predicted refusal starts the step at the chosen layer.
- autoRoute: virtual "Otomatik" model; one decision per user turn picks
  the combo, tool steps of the same turn keep it.
- finishCheck: an answer that stops early goes back to the same model once.
- A layer whose window cannot hold the request is skipped.
- policy: stage rules from a Dify workflow (karar_politikasi.py), fetched
  in the background, validated, last good copy kept; turns never wait.
- karar_sor.py: verifierCheck for the dynamic-workflow acceptance step.
- Bridge version 19, IDE 1.0.0-beta.39.

Tests: 299 bridge tests, 36 electron tests; live run on a side bridge
(port 8797) with Celeris and the AITURK gateway for every stage.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-14 07:48:33 +03:00

419 lines
20 KiB
Python

"""Decision stages (14 Oct 2026): settings schema 2, turn-start decisions, layer window filter, finish check
and Dify rules. Every new stage is off by default; the tests turn them on explicitly."""
import ast
import copy
import hashlib
import io
import json
import tempfile
import time
import unittest
from pathlib import Path
from unittest.mock import Mock, patch
import jev_denetim
import jev_settings
import karar_politikasi
import model_gecis
import pencere
import yarim
from test_jev_denetim import Yanit, method, sse, user
SOURCE = Path(__file__).with_name("kopru.py").read_text(encoding="utf-8")
TREE = ast.parse(SOURCE)
def fonksiyon(name):
return next(n for n in TREE.body if isinstance(n, ast.FunctionDef) and n.name == name)
class Ev(unittest.TestCase):
"""A temporary bridge home with a settings file."""
def setUp(self):
self.tmp = tempfile.TemporaryDirectory()
self.addCleanup(self.tmp.cleanup)
self.home = Path(self.tmp.name)
p = patch.object(jev_settings, 'home', return_value=self.home)
p.start()
self.addCleanup(p.stop)
self.settings = copy.deepcopy(jev_settings.DEFAULTS)
def save(self, settings=None, **extra):
data = {'settings': settings if settings is not None else self.settings}
data.update(extra)
(self.home / 'jev-settings.json').write_text(json.dumps(data), encoding='utf-8')
class Ayarlar(Ev):
def test_a_schema_1_file_gets_every_new_stage_disabled(self):
eski = {k: v for k, v in self.settings.items() if k not in jev_settings.SECTIONS}
self.save(eski)
s, _ = jev_settings.read()
for bolum in ('prefilter', 'autoRoute', 'finishCheck', 'verifierCheck'):
self.assertFalse(s[bolum]['enabled'], bolum)
self.assertEqual(s['policy']['source'], 'local')
self.assertEqual(s['watchModels'], jev_settings.DEFAULTS['watchModels'])
def test_a_partial_section_keeps_written_values_and_fills_the_rest(self):
self.settings['prefilter'] = {'enabled': True, 'threshold': 0.8}
self.save()
p = jev_settings.read()[0]['prefilter']
self.assertEqual((p['enabled'], p['threshold']), (True, 0.8))
self.assertEqual(p['target'], jev_settings.DEFAULTS['prefilter']['target'])
def test_bad_values_are_rejected(self):
for bolum, alan, deger in (('prefilter', 'threshold', 1.5), ('prefilter', 'timeoutSeconds', 0),
('autoRoute', 'model', ''), ('finishCheck', 'note', ''),
('policy', 'endpoint', 'http://dify.example'), ('policy', 'source', 'web')):
with self.subTest(bolum=bolum, alan=alan):
s = copy.deepcopy(self.settings)
s[bolum][alan] = deger
self.save(s)
with self.assertRaises(ValueError):
jev_settings.read()
def test_automatic_model_cannot_target_itself(self):
self.settings['autoRoute']['targets']['zor'] = 'Otomatik'
self.save()
with self.assertRaises(ValueError):
jev_settings.read()
def test_watch_channel_is_an_exact_name_from_settings(self):
self.settings['watchModels'] = ['Chat-Model']
self.save()
self.assertTrue(jev_denetim.izleme_mi('Chat-Model'))
for ad in ('chat-model', 'Chat-Model ', 'ide/Chat-Model', 'Sesli-Hizli-Zeki'):
self.assertFalse(jev_denetim.izleme_mi(ad), ad)
class Politika(Ev):
def test_dify_rules_apply_only_when_the_source_is_dify_and_they_are_valid(self):
(self.home / jev_settings.POLICY_FILE).write_text(json.dumps(
{'ok': True, 'overrides': {'prefilter': {'enabled': True, 'target': 'Katman-3'}}}), encoding='utf-8')
self.save()
self.assertFalse(jev_settings.read()[0]['prefilter']['enabled']) # source local
self.settings['policy']['source'] = 'dify'
self.save()
p = jev_settings.read()[0]['prefilter']
self.assertEqual((p['enabled'], p['target']), (True, 'Katman-3'))
self.assertFalse(jev_settings.read(overlay=False)[0]['prefilter']['enabled'])
(self.home / jev_settings.POLICY_FILE).write_text(json.dumps(
{'ok': True, 'overrides': {'prefilter': {'threshold': 7}}}), encoding='utf-8')
self.assertFalse(jev_settings.read()[0]['prefilter']['enabled']) # invalid rules: local in force
def test_rules_cannot_change_the_chain_or_unknown_sections(self):
s = jev_settings.read(overlay=False)[0]
with self.assertRaises(ValueError):
karar_politikasi.ayikla({'exemptModels': {'x': 1}}, s)
with self.assertRaises(ValueError):
karar_politikasi.ayikla({'autoRoute': {'targets': {'zor': 'Otomatik'}}}, s)
with self.assertRaises(ValueError):
karar_politikasi.ayikla('[1, 2]', s)
self.assertEqual(karar_politikasi.ayikla('{"autoRoute": {"enabled": true}}', s), {'autoRoute': {'enabled': True}})
def test_fetch_reads_the_kurallar_output_and_never_raises(self):
s = jev_settings.read(overlay=False)[0]
def yanit(govde):
r = Mock()
r.__enter__ = lambda self_: self_
r.__exit__ = lambda *a: False
r.read = lambda n=-1: json.dumps(govde).encode()
return r
istekler = []
def urlopen(req, timeout):
istekler.append(req)
return yanit({'data': {'outputs': {'kurallar': {'finishCheck': {'enabled': True}}}}})
sonuc = karar_politikasi.cek(s, 'app-key', urlopen=urlopen)
self.assertEqual(sonuc, {'ok': True, 'overrides': {'finishCheck': {'enabled': True}}})
self.assertTrue(istekler[0].full_url.endswith('/workflows/run'))
self.assertEqual(istekler[0].get_header('Authorization'), 'Bearer app-key')
self.assertFalse(karar_politikasi.cek(s, '', urlopen=urlopen)['ok'])
self.assertFalse(karar_politikasi.cek(s, 'k', urlopen=Mock(side_effect=OSError('kapali')))['ok'])
bos = karar_politikasi.cek(s, 'k', urlopen=lambda req, timeout: yanit({'data': {'outputs': {}}}))
self.assertEqual(bos['ok'], False)
def test_a_failed_fetch_keeps_the_last_good_rules(self):
jev_settings.write_policy({'ok': True, 'overrides': {'finishCheck': {'enabled': True}}})
kayit = jev_settings.write_policy({'ok': False, 'error': 'Dify kapali'})
self.assertTrue(kayit['ok'])
self.assertEqual(kayit['overrides'], {'finishCheck': {'enabled': True}})
self.assertEqual(kayit['error'], 'Dify kapali')
self.settings['policy']['source'] = 'dify'
self.save()
self.assertTrue(jev_settings.read()[0]['finishCheck']['enabled'])
class OnKarar(Ev):
def gonder(self, cevap):
self.govdeler = []
def f(govde, key):
self.govdeler.append(govde)
if isinstance(cevap, Exception):
raise cevap
return {'answers': {k: {'noul': v} for k, v in cevap.items()}}
return f
def test_automatic_routing_picks_research_then_hard_then_simple(self):
self.settings['autoRoute']['enabled'] = True
self.save()
t = self.settings['autoRoute']['targets']
for puan, hedef in (({'zor': 0.9, 'arastirma': 0.8}, t['arastirma']), ({'zor': 0.9, 'arastirma': 0.1}, t['zor']),
({'zor': 0.1, 'arastirma': 0.1}, t['basit'])):
sonuc = jev_denetim.on_karar('istek', True, False, 'k', gonder=self.gonder(puan))
self.assertEqual(sonuc['hedef'], hedef)
self.assertEqual(set(self.govdeler[0]['questions']), {'zor', 'arastirma'})
self.assertNotIn('assistant_reply', self.govdeler[0]['state'])
def test_no_decision_falls_to_the_strong_target(self):
self.settings['autoRoute']['enabled'] = True
self.save()
sonuc = jev_denetim.on_karar('istek', True, False, 'k', gonder=self.gonder(OSError('yok')))
self.assertEqual((sonuc['hedef'], sonuc['ret']), (self.settings['autoRoute']['targets']['zor'], None))
def test_refusal_question_goes_in_the_same_call(self):
self.settings['autoRoute']['enabled'] = True
self.settings['prefilter']['enabled'] = True
self.save()
sonuc = jev_denetim.on_karar('istek', True, True, 'k', gonder=self.gonder({'zor': 0.1, 'arastirma': 0.1, 'ret': 0.9}))
self.assertEqual(len(self.govdeler), 1)
self.assertEqual(set(self.govdeler[0]['questions']), {'zor', 'arastirma', 'ret'})
self.assertTrue(sonuc['ret'])
def test_a_missing_answer_is_no_decision(self):
self.settings['prefilter']['enabled'] = True
self.save()
sonuc = jev_denetim.on_karar('istek', False, True, 'k', gonder=self.gonder({}))
self.assertIsNone(sonuc['ret'])
class TurBasi(Ev):
"""kopru.karar_asamasi through its own source."""
def setUp(self):
super().setUp()
self.kararlar = []
self.logs = []
sahte_denetim = Mock(wraps=jev_denetim)
sahte_denetim.anahtar = lambda: 'k'
sahte_denetim.on_karar = self.on_karar
sahte_denetim.muaf_mi = jev_denetim.muaf_mi
sahte_denetim.sonrakiler = jev_denetim.sonrakiler
sahte_denetim.son_istek = jev_denetim.son_istek
self.cevap = {'hedef': None, 'ret': None, 'puan': None}
sahte_denetim.son_istek_sirasi = jev_denetim.son_istek_sirasi
self.env = {'IDE_PREFIX': 'ide/', 'ON_KARAR': {}, 'ON_KARAR_SINIR': 3, 'jev_settings': jev_settings,
'jev_denetim': sahte_denetim, 'model_gecis': model_gecis, 'hashlib': hashlib, 'time': time,
'log': self.logs.append}
exec(compile(ast.Module(body=[fonksiyon('karar_asamasi')], type_ignores=[]), 'kopru.py', 'exec'), self.env)
self.karar = self.env['karar_asamasi']
def on_karar(self, istek, oto, ret, key):
self.kararlar.append((istek, oto, ret))
return dict(self.cevap)
def istek(self, model, *girdi):
return {'model': model, 'prompt_cache_key': 'sohbet-1', 'input': list(girdi)}
def test_everything_off_changes_nothing_and_asks_nothing(self):
self.save()
req = self.istek('ide/Otomatik', user('yap'))
self.assertEqual(self.karar(1, req, 'sohbet-1', 'ide/Otomatik'), 'ide/Otomatik')
self.assertEqual(self.kararlar, [])
def test_automatic_choice_is_kept_for_the_tool_steps_of_the_same_turn(self):
self.settings['autoRoute']['enabled'] = True
self.save()
self.cevap = {'hedef': 'Arastirma', 'ret': None, 'puan': {'zor': 0.2, 'arastirma': 0.9}}
req = self.istek('ide/Otomatik', user('kaynaklari karsilastir'))
self.assertEqual(self.karar(1, req, 'sohbet-1', 'ide/Otomatik'), 'ide/Arastirma')
self.assertEqual(req['model'], 'ide/Arastirma')
arac = {'type': 'function_call_output', 'call_id': 'c1', 'output': 'ok'}
req2 = self.istek('ide/Otomatik', user('kaynaklari karsilastir'), {'type': 'function_call', 'call_id': 'c1'}, arac)
self.assertEqual(self.karar(2, req2, 'sohbet-1', 'ide/Otomatik'), 'ide/Arastirma')
self.assertEqual(len(self.kararlar), 1) # same turn: no second question
req3 = self.istek('ide/Otomatik', user('kaynaklari karsilastir'), arac, user('simdi kodu duzelt'))
self.cevap = {'hedef': 'Ajan', 'ret': None, 'puan': {'zor': 0.9, 'arastirma': 0.1}}
self.assertEqual(self.karar(3, req3, 'sohbet-1', 'ide/Otomatik'), 'ide/Ajan')
self.assertEqual(len(self.kararlar), 2) # new user message: new decision
def test_predicted_refusal_starts_at_the_configured_layer(self):
self.settings['prefilter']['enabled'] = True
self.save()
self.cevap = {'hedef': None, 'ret': True, 'puan': {'ret': 0.9}}
req = self.istek('ide/codex-gpt-6-astra', user('sert bir roast yaz'))
self.assertEqual(self.karar(1, req, 'sohbet-1', 'ide/codex-gpt-6-astra'), 'ide/Katman-2')
self.assertEqual(self.kararlar, [('sert bir roast yaz', False, True)])
def test_exempt_and_later_layers_are_not_asked(self):
self.settings['prefilter']['enabled'] = True
self.save()
for model in ('ide/Chat-Model', 'ide/Katman-2', 'ide/Katman-3', 'codex/gpt-6-astra'):
req = self.istek(model, user('yap'))
self.assertEqual(self.karar(1, req, 'sohbet-1', model), model)
self.assertEqual(self.kararlar, [])
def test_no_refusal_or_no_decision_keeps_the_model(self):
self.settings['prefilter']['enabled'] = True
self.save()
for cevap in ({'hedef': None, 'ret': False, 'puan': {'ret': 0.1}}, {'hedef': None, 'ret': None, 'puan': None}):
self.env['ON_KARAR'].clear()
self.cevap = cevap
req = self.istek('ide/Ajan', user('yap'))
self.assertEqual(self.karar(1, req, 'sohbet-1', 'ide/Ajan'), 'ide/Ajan')
self.assertEqual(req['model'], 'ide/Ajan')
def test_master_switch_off_disables_every_stage(self):
self.settings['enabled'] = False
self.settings['prefilter']['enabled'] = True
self.settings['autoRoute']['enabled'] = True
self.save()
req = self.istek('ide/Otomatik', user('yap'))
self.assertEqual(self.karar(1, req, 'sohbet-1', 'ide/Otomatik'), 'ide/Otomatik')
self.assertEqual(self.kararlar, [])
def test_turn_memory_is_bounded(self):
self.settings['prefilter']['enabled'] = True
self.save()
self.cevap = {'hedef': None, 'ret': False, 'puan': {'ret': 0.1}}
for i in range(6):
self.karar(i, {'model': 'ide/Ajan', 'prompt_cache_key': f's{i}', 'input': [user('yap')]}, f's{i}', 'ide/Ajan')
self.assertLessEqual(len(self.env['ON_KARAR']), 3)
class KatmanPenceresi(unittest.TestCase):
def test_a_layer_too_small_for_the_request_is_skipped(self):
logs = []
env = {'IDE_PREFIX': 'ide/', 'IDE_BILGI': {'ide/Katman-2': {'pencere': 1000}, 'ide/Katman-3': {'pencere': 10 ** 6}},
'pencere': pencere, 'log': logs.append}
exec(compile(ast.Module(body=[fonksiyon('sigan_katmanlar')], type_ignores=[]), 'kopru.py', 'exec'), env)
req = {'input': [user('x' * 20000)]}
self.assertEqual(env['sigan_katmanlar'](['Katman-2', 'Katman-3', 'Bilinmeyen'], req, '[1]'), ['Katman-3', 'Bilinmeyen'])
self.assertIn('Katman-2', logs[0])
class BitisDenetimi(Ev):
"""_denetimli_akis with finishCheck: an answer that stops early goes back to the same model once."""
def setUp(self):
super().setUp()
self.settings['finishCheck']['enabled'] = True
self.save()
self.asked, self.logs, self.devamlar = [], [], []
self.puanlar = []
sahte = Mock()
sahte.anahtar = lambda: 'k'
sahte.sor = self.sor
sahte.karar = jev_denetim.karar
sahte.bitmedi_mi = jev_denetim.bitmedi_mi
sahte.hazirlik_mi = jev_denetim.hazirlik_mi
sahte.muaf_mi = jev_denetim.muaf_mi
sahte.hata_gecis = jev_denetim.hata_gecis
env = {'jev_denetim': sahte, 'yarim': yarim, 'time': time, 'http': __import__('http.client'),
'log': self.logs.append, 'RESP_HOP': set(), 'json': json}
adlar = ('_akis_basligi', '_denetimli_akis', '_stream')
exec(compile(ast.Module(body=[method(a) for a in adlar], type_ignores=[]), 'kopru.py', 'exec'), env)
self.h = type('H', (), {a: env[a] for a in adlar})()
self.h.wfile = io.BytesIO()
self.h.send_response = self.h.send_header = self.h.end_headers = Mock()
def sor(self, istek, metin, key, bitis=False):
self.asked.append((metin, bitis))
return self.puanlar.pop(0) if self.puanlar else {'ret': 0.0, 'sart': 0.0}
def devam(self, ad, onceki):
self.devamlar.append((ad, onceki))
return Yanit(sse('gercekten bitti, cikti: OK')), None, f'[{ad} devam]'
def calistir(self, body):
self.h._denetimli_akis(Yanit(body), '[1]', None, None, 'yap', ['Katman-1'], Mock(), hedef='Ajan', devam=self.devam)
return self.gonderilen()
def gonderilen(self):
raw, out = self.h.wfile.getvalue(), b''
while raw:
size, _, rest = raw.partition(b'\r\n')
n = int(size, 16)
if n == 0:
break
out += rest[:n]
raw = rest[n + 2:]
return out
def test_an_early_stop_goes_back_once_and_the_retry_is_delivered(self):
self.puanlar = [{'ret': 0.0, 'sart': 0.0, 'bitmedi': 0.9}, {'ret': 0.0, 'sart': 0.0, 'bitmedi': 0.9}]
out = self.calistir(sse('Bitti.'))
self.assertEqual(self.devamlar, [('Ajan', 'Bitti.')])
self.assertIn(b'gercekten bitti', out)
self.assertNotIn(b'"Bitti."', out)
self.assertEqual([b for _, b in self.asked], [True, False]) # the retry is not asked again
def test_a_finished_answer_is_delivered_unchanged(self):
self.puanlar = [{'ret': 0.0, 'sart': 0.0, 'bitmedi': 0.1}]
out = self.calistir(sse('Tamam, test ciktisi: 3 passed.'))
self.assertEqual(self.devamlar, [])
self.assertIn(b'3 passed', out)
def test_a_refusal_still_goes_down_the_chain_not_back(self):
self.puanlar = [{'ret': 0.9, 'sart': 0.0, 'bitmedi': 0.9}]
ac = Mock(return_value=(Yanit(sse('katman')), None, '[K1]'))
self.h._denetimli_akis(Yanit(sse('Yapamam.')), '[1]', None, None, 'yap', ['Katman-1'], ac, hedef='Ajan', devam=self.devam)
self.assertEqual(self.devamlar, [])
ac.assert_called_once_with('Katman-1')
def test_a_failed_retry_delivers_the_first_answer(self):
self.puanlar = [{'ret': 0.0, 'sart': 0.0, 'bitmedi': 0.9}]
def bozuk(ad, onceki):
raise OSError('acilamadi')
self.h._denetimli_akis(Yanit(sse('Bitti.')), '[1]', None, None, 'yap', [], Mock(), hedef='Ajan', devam=bozuk)
self.assertIn(b'Bitti.', self.gonderilen())
def test_without_the_stage_the_finish_question_is_not_sent(self):
govdeler = []
def gonder(govde, key):
govdeler.append(govde)
return {'answers': {'ret': {'noul': 0.1}, 'sart': {'noul': 0.1}}}
puan = jev_denetim.sor('i', 'c', 'k', gonder=gonder)
self.assertNotIn('bitmedi', govdeler[0]['questions'])
self.assertNotIn('bitmedi', puan)
puan = jev_denetim.sor('i', 'c', 'k', gonder=gonder, bitis=True)
self.assertIn('bitmedi', govdeler[1]['questions'])
self.assertEqual(set(puan), {'ret', 'sart'}) # unanswered extra question: main verdict stands
self.assertFalse(jev_denetim.bitmedi_mi(puan))
if __name__ == '__main__':
unittest.main()
class DogrulayiciElemesi(Ev):
def test_off_by_default_and_asked_only_when_enabled(self):
import io as _io
import karar_sor
self.save()
out = _io.StringIO()
self.assertEqual(karar_sor.main(['dogrulayici'], _io.StringIO('{"user_request": "x", "assistant_reply": "kabul"}'), out), 0)
self.assertEqual(json.loads(out.getvalue())['enabled'], False)
self.settings['verifierCheck']['enabled'] = True
self.save()
govdeler = []
def gonder(govde, key):
govdeler.append(govde)
return {'answers': {'kanitsiz': {'noul': 0.92}}}
sonuc = karar_sor.dogrulayici({'user_request': 'olcut', 'assistant_reply': '{"karar":"kabul","bulgular":[]}'}, 'k', gonder)
self.assertEqual((sonuc['enabled'], sonuc['kanitsiz'], sonuc['puan']), (True, True, 0.92))
self.assertEqual(set(govdeler[0]['questions']), {'kanitsiz'})
bozuk = karar_sor.dogrulayici({'user_request': 'o', 'assistant_reply': 'r'}, 'k', Mock(side_effect=OSError('yok')))
self.assertIsNone(bozuk['kanitsiz'])
def test_wrong_usage_exits_2(self):
import io as _io
import karar_sor
self.assertEqual(karar_sor.main(['baska'], _io.StringIO('{}'), _io.StringIO()), 2)