feat(governance): 新增 post-write verifier package
This commit is contained in:
@@ -13,9 +13,9 @@ def test_load_latest_ai_agent_interaction_learning_proof_reads_committed_snapsho
|
||||
data = load_latest_ai_agent_interaction_learning_proof()
|
||||
|
||||
assert data["schema_version"] == "ai_agent_interaction_learning_proof_v1"
|
||||
assert data["program_status"]["overall_completion_percent"] == 94
|
||||
assert data["program_status"]["current_task_id"] == "P2-403G"
|
||||
assert data["program_status"]["next_task_id"] == "P2-403H"
|
||||
assert data["program_status"]["overall_completion_percent"] == 97
|
||||
assert data["program_status"]["current_task_id"] == "P2-403H"
|
||||
assert data["program_status"]["next_task_id"] == "P2-403I"
|
||||
assert data["program_status"]["read_only_mode"] is True
|
||||
assert data["program_status"]["runtime_authority"] == "proof_surface_only_no_live_worker"
|
||||
assert data["live_truth"]["runtime_loop_enabled"] is False
|
||||
|
||||
@@ -16,9 +16,9 @@ def test_ai_agent_interaction_learning_proof_endpoint_returns_committed_snapshot
|
||||
assert response.status_code == 200
|
||||
data = response.json()
|
||||
assert data["schema_version"] == "ai_agent_interaction_learning_proof_v1"
|
||||
assert data["program_status"]["overall_completion_percent"] == 94
|
||||
assert data["program_status"]["current_task_id"] == "P2-403G"
|
||||
assert data["program_status"]["next_task_id"] == "P2-403H"
|
||||
assert data["program_status"]["overall_completion_percent"] == 97
|
||||
assert data["program_status"]["current_task_id"] == "P2-403H"
|
||||
assert data["program_status"]["next_task_id"] == "P2-403I"
|
||||
assert data["program_status"]["read_only_mode"] is True
|
||||
assert data["live_truth"]["runtime_loop_enabled"] is False
|
||||
assert data["live_truth"]["active_live_agent_sessions"] == 0
|
||||
|
||||
74
apps/api/tests/test_ai_agent_post_write_verifier_package.py
Normal file
74
apps/api/tests/test_ai_agent_post_write_verifier_package.py
Normal file
@@ -0,0 +1,74 @@
|
||||
import copy
|
||||
import json
|
||||
|
||||
import pytest
|
||||
|
||||
from src.services.ai_agent_post_write_verifier_package import (
|
||||
load_latest_ai_agent_post_write_verifier_package,
|
||||
)
|
||||
|
||||
|
||||
def _write_snapshot(tmp_path, payload):
|
||||
path = tmp_path / "ai_agent_post_write_verifier_package_2026-06-12.json"
|
||||
path.write_text(json.dumps(payload), encoding="utf-8")
|
||||
return path
|
||||
|
||||
|
||||
def test_load_latest_ai_agent_post_write_verifier_package():
|
||||
data = load_latest_ai_agent_post_write_verifier_package()
|
||||
|
||||
assert data["schema_version"] == "ai_agent_post_write_verifier_package_v1"
|
||||
assert data["program_status"]["current_task_id"] == "P2-403H"
|
||||
assert data["program_status"]["next_task_id"] == "P2-403I"
|
||||
assert data["program_status"]["overall_completion_percent"] == 97
|
||||
assert data["verifier_truth"]["runtime_write_allowed"] is False
|
||||
assert data["verifier_truth"]["post_write_verifier_implemented"] is False
|
||||
assert data["verifier_truth"]["post_write_verifier_executed_count"] == 0
|
||||
assert data["verifier_truth"]["rollback_work_item_created_count"] == 0
|
||||
assert data["rollups"]["verification_target_count"] == len(data["verification_targets"])
|
||||
assert data["rollups"]["failure_lane_count"] == len(data["failure_lanes"])
|
||||
assert data["rollups"]["operator_action_count"] == len(data["operator_actions"])
|
||||
assert data["rollups"]["live_verifier_execution_count"] == 0
|
||||
|
||||
|
||||
def test_rejects_post_write_verifier_implemented(tmp_path):
|
||||
data = load_latest_ai_agent_post_write_verifier_package()
|
||||
bad = copy.deepcopy(data)
|
||||
bad["verifier_truth"]["post_write_verifier_implemented"] = True
|
||||
_write_snapshot(tmp_path, bad)
|
||||
|
||||
with pytest.raises(ValueError, match="verifier runtime flags"):
|
||||
load_latest_ai_agent_post_write_verifier_package(tmp_path)
|
||||
|
||||
|
||||
def test_rejects_rollback_count_increment(tmp_path):
|
||||
data = load_latest_ai_agent_post_write_verifier_package()
|
||||
bad = copy.deepcopy(data)
|
||||
bad["verifier_truth"]["rollback_work_item_created_count"] = 1
|
||||
_write_snapshot(tmp_path, bad)
|
||||
|
||||
with pytest.raises(ValueError, match="verifier counts"):
|
||||
load_latest_ai_agent_post_write_verifier_package(tmp_path)
|
||||
|
||||
|
||||
def test_rejects_missing_required_input(tmp_path):
|
||||
data = load_latest_ai_agent_post_write_verifier_package()
|
||||
bad = copy.deepcopy(data)
|
||||
bad["verifier_package"]["required_inputs"] = [
|
||||
field for field in bad["verifier_package"]["required_inputs"] if field != "canonical_readback_query"
|
||||
]
|
||||
bad["rollups"]["required_input_count"] = len(bad["verifier_package"]["required_inputs"])
|
||||
_write_snapshot(tmp_path, bad)
|
||||
|
||||
with pytest.raises(ValueError, match="missing required inputs"):
|
||||
load_latest_ai_agent_post_write_verifier_package(tmp_path)
|
||||
|
||||
|
||||
def test_rejects_rollup_mismatch(tmp_path):
|
||||
data = load_latest_ai_agent_post_write_verifier_package()
|
||||
bad = copy.deepcopy(data)
|
||||
bad["rollups"]["failure_lane_count"] = 999
|
||||
_write_snapshot(tmp_path, bad)
|
||||
|
||||
with pytest.raises(ValueError, match="rollup counts"):
|
||||
load_latest_ai_agent_post_write_verifier_package(tmp_path)
|
||||
@@ -0,0 +1,23 @@
|
||||
from fastapi.testclient import TestClient
|
||||
|
||||
from src.main import app
|
||||
|
||||
|
||||
def test_get_ai_agent_post_write_verifier_package_api():
|
||||
client = TestClient(app)
|
||||
response = client.get("/api/v1/agents/agent-post-write-verifier-package")
|
||||
|
||||
assert response.status_code == 200
|
||||
data = response.json()
|
||||
assert data["schema_version"] == "ai_agent_post_write_verifier_package_v1"
|
||||
assert data["program_status"]["current_task_id"] == "P2-403H"
|
||||
assert data["program_status"]["next_task_id"] == "P2-403I"
|
||||
assert data["program_status"]["overall_completion_percent"] == 97
|
||||
assert data["verifier_truth"]["runtime_write_allowed"] is False
|
||||
assert data["verifier_truth"]["post_write_verifier_implemented"] is False
|
||||
assert data["verifier_truth"]["post_write_verifier_executed_count"] == 0
|
||||
assert data["verifier_truth"]["rollback_work_item_created_count"] == 0
|
||||
assert data["rollups"]["verification_target_count"] == 4
|
||||
assert data["rollups"]["failure_lane_count"] == 3
|
||||
assert data["rollups"]["operator_action_count"] == 4
|
||||
assert data["rollups"]["live_verifier_execution_count"] == 0
|
||||
@@ -14,8 +14,8 @@ def test_load_latest_ai_agent_proactive_operations_contract_reads_committed_snap
|
||||
|
||||
assert data["schema_version"] == "ai_agent_proactive_operations_contract_v1"
|
||||
assert data["program_status"]["overall_completion_percent"] == 100
|
||||
assert data["program_status"]["current_task_id"] == "P2-403G"
|
||||
assert data["program_status"]["next_task_id"] == "P2-403H"
|
||||
assert data["program_status"]["current_task_id"] == "P2-403H"
|
||||
assert data["program_status"]["next_task_id"] == "P2-403I"
|
||||
assert data["program_status"]["read_only_mode"] is True
|
||||
assert data["program_status"]["runtime_authority"] == "contract_only_no_version_or_runtime_update"
|
||||
assert data["approval_boundaries"]["runtime_version_update_allowed"] is False
|
||||
@@ -25,7 +25,7 @@ def test_load_latest_ai_agent_proactive_operations_contract_reads_committed_snap
|
||||
assert data["approval_boundaries"]["telegram_direct_send_allowed"] is False
|
||||
assert data["rollups"]["version_domain_count"] == len(data["version_lifecycle_domains"]) == 12
|
||||
assert data["rollups"]["delegable_capability_count"] == len(data["delegable_capabilities"]) == 24
|
||||
assert data["rollups"]["rollout_task_count"] == len(data["rollout_tasks"]) == 14
|
||||
assert data["rollups"]["rollout_task_count"] == len(data["rollout_tasks"]) == 15
|
||||
assert data["rollups"]["auto_execute_allowed_count"] == 0
|
||||
assert any(domain["domain_id"] == "ai_agents_models" for domain in data["version_lifecycle_domains"])
|
||||
assert any(
|
||||
|
||||
@@ -17,8 +17,8 @@ def test_ai_agent_proactive_operations_contract_endpoint_returns_committed_snaps
|
||||
data = response.json()
|
||||
assert data["schema_version"] == "ai_agent_proactive_operations_contract_v1"
|
||||
assert data["program_status"]["overall_completion_percent"] == 100
|
||||
assert data["program_status"]["current_task_id"] == "P2-403G"
|
||||
assert data["program_status"]["next_task_id"] == "P2-403H"
|
||||
assert data["program_status"]["current_task_id"] == "P2-403H"
|
||||
assert data["program_status"]["next_task_id"] == "P2-403I"
|
||||
assert data["program_status"]["read_only_mode"] is True
|
||||
assert data["approval_boundaries"]["runtime_version_update_allowed"] is False
|
||||
assert data["approval_boundaries"]["package_upgrade_allowed"] is False
|
||||
@@ -26,7 +26,7 @@ def test_ai_agent_proactive_operations_contract_endpoint_returns_committed_snaps
|
||||
assert data["approval_boundaries"]["telegram_direct_send_allowed"] is False
|
||||
assert data["rollups"]["version_domain_count"] == 12
|
||||
assert data["rollups"]["delegable_capability_count"] == 24
|
||||
assert data["rollups"]["rollout_task_count"] == 14
|
||||
assert data["rollups"]["rollout_task_count"] == 15
|
||||
assert data["rollups"]["auto_execute_allowed_count"] == 0
|
||||
assert any(domain["domain_id"] == "host_os_packages" for domain in data["version_lifecycle_domains"])
|
||||
assert any(
|
||||
|
||||
Reference in New Issue
Block a user