feat(ai): 新增 P2-408 中低風險白名單

apps/api/tests/test_ai_agent_low_medium_risk_whitelist.py

@@ -0,0 +1,142 @@
+from __future__ import annotations
+
+import copy
+import json
+from pathlib import Path
+
+import pytest
+
+from src.services.ai_agent_low_medium_risk_whitelist import (
+    load_latest_ai_agent_low_medium_risk_whitelist,
+)
+
+_REPO_ROOT = Path(__file__).resolve().parents[3]
+_COMMITTED_SNAPSHOT = (
+    _REPO_ROOT
+    / "docs"
+    / "evaluations"
+    / "ai_agent_low_medium_risk_whitelist_2026-06-18.json"
+)
+
+
+def test_load_latest_ai_agent_low_medium_risk_whitelist_reads_newest_file(tmp_path):
+    older = _snapshot(generated_at="2026-06-17T00:00:00+08:00")
+    newer = _snapshot(generated_at="2026-06-18T18:42:00+08:00")
+    (tmp_path / "ai_agent_low_medium_risk_whitelist_2026-06-17.json").write_text(
+        json.dumps(older),
+        encoding="utf-8",
+    )
+    (tmp_path / "ai_agent_low_medium_risk_whitelist_2026-06-18.json").write_text(
+        json.dumps(newer),
+        encoding="utf-8",
+    )
+
+    loaded = load_latest_ai_agent_low_medium_risk_whitelist(tmp_path)
+
+    assert loaded["generated_at"] == "2026-06-18T18:42:00+08:00"
+    assert loaded["program_status"]["current_task_id"] == "P2-408"
+    assert loaded["program_status"]["next_task_id"] == "P2-409"
+    assert loaded["program_status"]["read_only_mode"] is True
+    assert loaded["rollups"]["low_risk_candidate_count"] == 3
+    assert loaded["rollups"]["medium_risk_candidate_count"] == 3
+    assert loaded["rollups"]["auto_worker_run_count"] == 0
+
+
+def test_ai_agent_low_medium_risk_whitelist_requires_read_only_mode(tmp_path):
+    snapshot = _snapshot()
+    snapshot["program_status"]["read_only_mode"] = False
+    _write_snapshot(tmp_path, snapshot)
+
+    with pytest.raises(ValueError, match="program_status"):
+        load_latest_ai_agent_low_medium_risk_whitelist(tmp_path)
+
+
+def test_ai_agent_low_medium_risk_whitelist_blocks_live_execution(tmp_path):
+    snapshot = _snapshot()
+    snapshot["whitelist_candidates"][0]["live_execution_allowed"] = True
+    _write_snapshot(tmp_path, snapshot)
+
+    with pytest.raises(ValueError, match="live_execution_allowed"):
+        load_latest_ai_agent_low_medium_risk_whitelist(tmp_path)
+
+
+def test_ai_agent_low_medium_risk_whitelist_blocks_gateway_queue_write(tmp_path):
+    snapshot = _snapshot()
+    snapshot["activation_boundaries"]["gateway_queue_write_enabled"] = True
+    _write_snapshot(tmp_path, snapshot)
+
+    with pytest.raises(ValueError, match="activation boundaries"):
+        load_latest_ai_agent_low_medium_risk_whitelist(tmp_path)
+
+
+def test_ai_agent_low_medium_risk_whitelist_requires_dry_run_verifier(tmp_path):
+    snapshot = _snapshot()
+    snapshot["whitelist_candidates"][0]["dry_run_verifier_id"] = "missing_verifier"
+    _write_snapshot(tmp_path, snapshot)
+
+    with pytest.raises(ValueError, match="missing dry-run verifiers"):
+        load_latest_ai_agent_low_medium_risk_whitelist(tmp_path)
+
+
+def test_ai_agent_low_medium_risk_whitelist_requires_rollback_proof(tmp_path):
+    snapshot = _snapshot()
+    snapshot["whitelist_candidates"][0]["rollback_proof_id"] = "missing_rollback"
+    _write_snapshot(tmp_path, snapshot)
+
+    with pytest.raises(ValueError, match="missing rollback proofs"):
+        load_latest_ai_agent_low_medium_risk_whitelist(tmp_path)
+
+
+def test_ai_agent_low_medium_risk_whitelist_requires_audit_reason_template(tmp_path):
+    snapshot = _snapshot()
+    snapshot["whitelist_candidates"][0]["audit_reason_template_id"] = "missing_audit_template"
+    _write_snapshot(tmp_path, snapshot)
+
+    with pytest.raises(ValueError, match="missing audit reason templates"):
+        load_latest_ai_agent_low_medium_risk_whitelist(tmp_path)
+
+
+def test_ai_agent_low_medium_risk_whitelist_rejects_high_risk_candidate(tmp_path):
+    snapshot = _snapshot()
+    snapshot["whitelist_candidates"][0]["risk_tier"] = "high"
+    _write_snapshot(tmp_path, snapshot)
+
+    with pytest.raises(ValueError, match="must be low or medium"):
+        load_latest_ai_agent_low_medium_risk_whitelist(tmp_path)
+
+
+def test_ai_agent_low_medium_risk_whitelist_requires_rollup_consistency(tmp_path):
+    snapshot = _snapshot()
+    snapshot["rollups"]["whitelist_candidate_count"] = 99
+    _write_snapshot(tmp_path, snapshot)
+
+    with pytest.raises(ValueError, match="rollup counts"):
+        load_latest_ai_agent_low_medium_risk_whitelist(tmp_path)
+
+
+def test_ai_agent_low_medium_risk_whitelist_rejects_private_terms(tmp_path):
+    snapshot = _snapshot()
+    snapshot["whitelist_candidates"][0]["allowed_no_write_outputs"] = ["請把 In app browser 內容放進前端"]
+    _write_snapshot(tmp_path, snapshot)
+
+    with pytest.raises(ValueError, match="forbidden public terms"):
+        load_latest_ai_agent_low_medium_risk_whitelist(tmp_path)
+
+
+def test_ai_agent_low_medium_risk_whitelist_fails_when_missing(tmp_path):
+    with pytest.raises(FileNotFoundError):
+        load_latest_ai_agent_low_medium_risk_whitelist(tmp_path)
+
+
+def _snapshot(*, generated_at: str = "2026-06-18T18:42:00+08:00") -> dict:
+    payload = json.loads(_COMMITTED_SNAPSHOT.read_text(encoding="utf-8"))
+    cloned = copy.deepcopy(payload)
+    cloned["generated_at"] = generated_at
+    return cloned
+
+
+def _write_snapshot(tmp_path, payload: dict) -> None:
+    (tmp_path / "ai_agent_low_medium_risk_whitelist_2026-06-18.json").write_text(
+        json.dumps(payload),
+        encoding="utf-8",
+    )

apps/api/tests/test_ai_agent_low_medium_risk_whitelist_api.py

@@ -0,0 +1,58 @@
+from __future__ import annotations
+
+from fastapi import FastAPI
+from fastapi.testclient import TestClient
+
+from src.api.v1.agents import router
+
+
+def test_ai_agent_low_medium_risk_whitelist_endpoint_returns_committed_snapshot():
+    app = FastAPI()
+    app.include_router(router, prefix="/api/v1")
+    client = TestClient(app)
+
+    response = client.get("/api/v1/agents/agent-low-medium-risk-whitelist")
+
+    assert response.status_code == 200
+    data = response.json()
+    assert data["schema_version"] == "ai_agent_low_medium_risk_whitelist_v1"
+    assert data["program_status"]["current_task_id"] == "P2-408"
+    assert data["program_status"]["next_task_id"] == "P2-409"
+    assert data["program_status"]["read_only_mode"] is True
+    assert (
+        data["program_status"]["runtime_authority"]
+        == "low_medium_risk_whitelist_no_live_execution_committed_snapshot"
+    )
+    assert data["rollups"]["source_readback_count"] == len(data["source_readbacks"]) == 5
+    assert data["rollups"]["whitelist_candidate_count"] == len(data["whitelist_candidates"]) == 6
+    assert data["rollups"]["low_risk_candidate_count"] == 3
+    assert data["rollups"]["medium_risk_candidate_count"] == 3
+    assert data["rollups"]["dry_run_verifier_count"] == len(data["dry_run_verifiers"]) == 5
+    assert data["rollups"]["rollback_proof_count"] == len(data["rollback_proofs"]) == 5
+    assert data["rollups"]["audit_reason_template_count"] == len(data["audit_reason_templates"]) == 6
+    assert data["rollups"]["high_risk_redirect_count"] == len(data["high_risk_redirects"]) == 3
+    assert data["rollups"]["owner_review_gate_count"] == len(data["owner_review_gates"]) == 3
+    assert data["rollups"]["live_execution_approval_required_count"] == 6
+    assert data["rollups"]["blocked_runtime_action_count"] == 27
+    assert data["rollups"]["auto_worker_run_count"] == 0
+    assert data["rollups"]["low_risk_execution_count"] == 0
+    assert data["rollups"]["medium_risk_execution_count"] == 0
+    assert data["rollups"]["gateway_queue_write_count"] == 0
+    assert data["rollups"]["telegram_send_count"] == 0
+    assert data["rollups"]["bot_api_call_count"] == 0
+    assert data["rollups"]["receipt_production_write_count"] == 0
+    assert data["rollups"]["production_write_count"] == 0
+    assert data["rollups"]["secret_read_count"] == 0
+    assert data["rollups"]["paid_api_call_count"] == 0
+    assert data["rollups"]["host_write_count"] == 0
+    assert data["rollups"]["kubectl_action_count"] == 0
+    assert data["rollups"]["destructive_operation_count"] == 0
+    assert data["telegram_policy"]["canonical_room"] == "AwoooI SRE 戰情室"
+    assert data["telegram_policy"]["canonical_room_env"] == "SRE_GROUP_CHAT_ID"
+    assert data["telegram_policy"]["telegram_send_allowed"] is False
+    assert data["telegram_policy"]["gateway_queue_write_allowed"] is False
+    assert data["telegram_policy"]["direct_bot_api_allowed"] is False
+    assert data["activation_boundaries"]["auto_worker_enabled"] is False
+    assert data["activation_boundaries"]["low_risk_live_execution_enabled"] is False
+    assert data["activation_boundaries"]["medium_risk_live_execution_enabled"] is False
+    assert data["activation_boundaries"]["openclaw_replacement_allowed"] is False