The agent that grows with you
Python aiai-agentai-agentsanthropicchatgptclaudeclaude-codecodexhermeshermes-agentllmnous-researchopenai
"""Worker receipts cover the real transcript serializer, not a text-only substitute."""
import pytest
from hermes_state import SessionDB
from hermes_state_runtime import RuntimeStoreError, begin_runtime_epoch, register_worker_execution
from agent.message_metadata import CANONICAL_ROW, DB_ROW_SNAPSHOT
def test_structured_worker_receipts_are_atomic_and_scoped(tmp_path):
import hermes_state_runtime as runtime
assert callable(getattr(runtime, 'mutate_worker_execution', None))
db = SessionDB(tmp_path / 'state.db')
try:
db.create_session('assigned', source='cli')
db.create_session('foreign', source='cli')
epoch = begin_runtime_epoch(db, instance_id='owner')
register_worker_execution(db, epoch=epoch, execution_id='worker', session_id='assigned',
generation=0, kind='compute', adoption_secret='private')
scope = dict(epoch=epoch, execution_id='worker', session_id='assigned', generation=0)
def apply(seq, op, payload, **overrides):
return runtime.mutate_worker_execution(db, **(scope | overrides), sequence=seq,
operation=op, payload=payload)
assert apply(1, 'turn.acquire', {'holder': 'worker-lease', 'ttl_seconds': 300})['value']
messages = [
{'role': 'user', 'content': [{'type': 'text', 'text': 'hello\ud800'}], 'api_content': 'wire'},
{'role': 'assistant', 'content': None, 'reasoning': 'private reason',
'tool_calls': [{'id': 'call1', 'type': 'function', 'function': {'name': 'terminal', 'arguments': '{}'}}]},
{'role': 'tool', 'content': 'tool-marker', 'tool_call_id': 'call1', 'tool_name': 'terminal'},
]
payload = {'messages': messages, 'turn_lease_holder': 'worker-lease'}
receipt = apply(2, 'transcript.append', payload)
assert receipt['count'] == 3
assert all(m['_row_id'] > 0 for m in receipt['annotations'])
assert apply(2, 'transcript.append', payload) == receipt
with pytest.raises(RuntimeStoreError, match='admission_conflict'):
apply(2, 'transcript.append', {'messages': [{'role': 'user', 'content': 'different'}]})
with pytest.raises(RuntimeStoreError, match='permission_denied'):
apply(3, 'transcript.append', payload, session_id='foreign')
with pytest.raises(RuntimeStoreError, match='stale_generation'):
apply(3, 'transcript.append', payload, generation=1)
assert len(db.get_messages('assigned')) == 3
stored = db.get_messages('assigned')
assert stored[1]['reasoning'] == 'private reason'
assert stored[2]['tool_call_id'] == 'call1'
assert stored[2]['content'] == 'tool-marker'
with pytest.raises(RuntimeStoreError, match='invalid_params'):
apply(3, 'SQL', {'sql': 'DELETE FROM messages'})
assert apply(3, 'turn.renew', {'holder': 'worker-lease', 'ttl_seconds': 300})['value']
assert apply(4, 'turn.release', {'holder': 'worker-lease'}) == {'value': None}
assistant = {'role': 'assistant', 'content': [{'type': 'text', 'text': 'canonical winner\ud800'}]}
winner = apply(5, 'transcript.append', {'messages': [assistant]})
repaired = apply(6, 'transcript.append', {'messages': [{'role': 'assistant',
'content': 'loser', '_row_id': winner['annotations'][0]['_row_id']}]})
assert repaired['count'] == 0
adopted = repaired['annotations'][0]
assert adopted[CANONICAL_ROW]['content'] == assistant['content']
assert adopted[DB_ROW_SNAPSHOT] and adopted['_row_id'] == winner['annotations'][0]['_row_id']
terminal = apply(7, 'execution.finish', {})
assert terminal['status'] == 'terminal'
assert apply(7, 'execution.finish', {}) == terminal
with pytest.raises(RuntimeStoreError, match='stale_generation'):
apply(8, 'usage.main', {'input_tokens': 1})
finally:
db.close()