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
"""
AiSOC Detection Pack v1 - Combined Specification Index
======================================================
Imports the rule tables from `detection_specs.py`, `detection_specs_part2.py`,
and the `detection_specs_part3_*.py` expansion modules, then exposes a single
`CATEGORIES` mapping plus an `all_specs()` iterator.
The category names map directly onto subdirectories under `detections/`.
Tier mapping:
* `detection_specs.py` + `detection_specs_part2.py` — original 200 hand-curated
native specs.
* `detection_specs_part3_*.py` — 600 additional native specs generated through
the compact `S()` builder (see `detection_specs_part3_helpers.py`). These
follow the same fixture + MITRE quality gate as the originals.
* `detection_specs_part4_threat_anomaly.py` — depth plan 3.5. Threat
conditions paired with depth plan 3.4 baseline scores. Fixtures are
hand-authored rather than synthesised, because the content of these
rules is a relationship and a fixture copied from the clause cannot
test it.
"""
from __future__ import annotations
from collections.abc import Iterable
# Local imports must be relative to scripts/ (that is the cwd when invoked).
from detection_specs import CLOUD, IDENTITY # type: ignore[import-not-found]
from detection_specs_part2 import ( # type: ignore[import-not-found]
APPLICATION,
DATA_EXFIL,
ENDPOINT,
NETWORK,
)
from detection_specs_part3_application import APPLICATION_EXTRA # type: ignore[import-not-found]
from detection_specs_part3_cloud import CLOUD_EXTRA # type: ignore[import-not-found]
from detection_specs_part3_endpoint import ENDPOINT_EXTRA # type: ignore[import-not-found]
from detection_specs_part3_identity import IDENTITY_EXTRA # type: ignore[import-not-found]
from detection_specs_part3_network import NETWORK_EXTRA # type: ignore[import-not-found]
from detection_specs_part4_threat_anomaly import THREAT_ANOMALY # type: ignore[import-not-found]
CATEGORIES: dict[str, list[dict]] = {
"cloud": [*CLOUD, *CLOUD_EXTRA],
"identity": [*IDENTITY, *IDENTITY_EXTRA],
"endpoint": [*ENDPOINT, *ENDPOINT_EXTRA],
"network": [*NETWORK, *NETWORK_EXTRA],
"application": [*APPLICATION, *APPLICATION_EXTRA],
"data-exfil": DATA_EXFIL,
# Depth plan 3.5. Its own category because the claim is different:
# these pair a threat condition with a depth plan 3.4 baseline score,
# so a fresh install has the rule before it has the signal. Counted
# separately in scripts/project_stats.py for the same reason.
"threat-anomaly": THREAT_ANOMALY,
}
def all_specs() -> Iterable[tuple[str, dict]]:
"""Yield (category, spec) tuples in deterministic order."""
for category in sorted(CATEGORIES.keys()):
for spec in CATEGORIES[category]:
yield category, spec
def total_counts() -> dict[str, int]:
return {category: len(rules) for category, rules in CATEGORIES.items()}
if __name__ == "__main__":
counts = total_counts()
total = sum(counts.values())
print("AiSOC Detection Pack v1 spec counts:")
for category, count in sorted(counts.items()):
print(f" {category:14s} {count:4d}")
print(f" {'TOTAL':14s} {total:4d}")