Open-source AI Security Operations Center: alert fusion, LLM-agent triage, MITRE ATT&CK investigation, and a replayable decision ledger for every agent step. Self-hostable, runs with no API keys, MIT licensed. Ships an MCP server for Claude, Cursor and Continue.
Python agentic-aiai-securityalert-triageblue-teamcybersecuritydetection-engineeringincident-responselanggraphllm-agentsmcpmitre-attackmodel-context-protocolpurple-teamsecurity-operationsself-hostedsiemsoarsocthreat-detectionthreat-intelligence
"""Wave 2 - detection backtesting over the lake (pure logic).
The lake-fetch path needs a live ClickHouse and is covered by the integration
gate; these tests pin the pure, deterministic pieces: bounded/allow-listed SQL
construction, row->event mapping (with JSON payload expansion), and the
would-fire count/hit-rate/sample summary.
"""
from __future__ import annotations
from app.services.backtest import (
backtest_rule,
build_backtest_sql,
rows_to_events,
sanitize_source,
)
def test_build_backtest_sql_bounds_and_source():
sql = build_backtest_sql(window_days=30, limit=5000, source="okta_system_log")
assert "FROM aisoc.raw_events" in sql
assert "INTERVAL 30 DAY" in sql
assert "connector_type = 'okta_system_log'" in sql
assert "LIMIT 5000" in sql
# window + limit are clamped to sane maxima
assert "INTERVAL 365 DAY" in build_backtest_sql(window_days=9999, limit=1, source=None)
assert "LIMIT 100000" in build_backtest_sql(window_days=1, limit=10**9, source=None)
def test_source_filter_is_injection_sanitised():
sql = build_backtest_sql(window_days=1, limit=1, source="okta'; DROP TABLE x;--")
assert "DROP TABLE" not in sql
assert sanitize_source("Okta_System-Log.1") == "okta_system-log.1"
assert sanitize_source("'; drop") == "drop"
assert sanitize_source("!!!") is None
assert sanitize_source(None) is None
def test_rows_to_events_zip_and_expand_payload():
cols = ["event_id", "connector_type", "raw_data"]
rows = [["e1", "okta", '{"user": "ceo", "action": "login"}']]
events = rows_to_events(cols, rows)
assert events[0]["connector_type"] == "okta"
assert events[0]["user"] == "ceo" # expanded from the JSON payload column
assert events[0]["action"] == "login"
def test_backtest_rule_counts_hit_rate_and_samples():
events = [{"msg": "malware detected"}, {"msg": "benign login"}, {"msg": "more malware here"}]
result = backtest_rule(rule_language="regex", rule_body="malware", events=events, sample_limit=1)
assert result["events_scanned"] == 3
assert result["would_fire"] == 2
assert result["hit_rate"] == round(2 / 3, 4)
assert len(result["sample_matches"]) == 1 # capped to sample_limit
assert result["error"] is None
def test_backtest_unsupported_language_reports_error_no_fire():
result = backtest_rule(rule_language="esql", rule_body="x", events=[{"a": 1}])
assert result["would_fire"] == 0
assert result["error"] is not None
def test_backtest_empty_lake_is_zero_not_crash():
result = backtest_rule(rule_language="regex", rule_body="anything", events=[])
assert result["events_scanned"] == 0
assert result["would_fire"] == 0
assert result["hit_rate"] == 0.0