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
"""Unit tests for the agent tools surface (Workstream 4).
These tests cover the *pure* helpers that shape ``GET /api/v1/agents/tools``
output — group lookup, description fallback, and capability normalisation
across both list-of-strings and list-of-dicts catalog payloads. The
endpoint itself is covered by integration tests once the auth fixtures
land; the helpers are the parts most likely to silently misbehave when
the connectors microservice changes its catalog wire format.
"""
from __future__ import annotations
from app.api.v1.endpoints.agents import (
_CAPABILITY_GROUP_LOOKUP,
_capability_descriptions,
_capability_group_of,
_default_description,
)
# ----------------------------------------------------------- group lookup
def test_capability_group_lookup_covers_taxonomy() -> None:
"""Every documented capability bucket must have at least one entry.
If someone adds a new capability bucket and forgets to wire it into
the lookup, we'd silently surface every verb in that bucket as
'unknown'. This catches that drift before it ships.
"""
expected_groups = {
"read",
"query",
"pivot",
"enrich",
"contain",
"remediate",
"ticket",
"audit",
}
seen_groups = set(_CAPABILITY_GROUP_LOOKUP.values())
assert expected_groups <= seen_groups
def test_capability_group_of_known_value() -> None:
assert _capability_group_of("isolate_host") == "contain"
assert _capability_group_of("pull_alerts") == "read"
assert _capability_group_of("push_case") == "ticket"
def test_capability_group_of_unknown_value_is_soft_signal() -> None:
"""Unknown capabilities surface as 'unknown' rather than raising.
Dropping a verb the connector class declared would silently hide
functionality from the agent. 'unknown' is a soft drift signal — the
agent layer can log a warning and decide whether to use it anyway.
"""
assert _capability_group_of("future_capability_we_havent_added_yet") == "unknown"
# ----------------------------------------------------------- descriptions
def test_default_description_humanises_underscores() -> None:
assert _default_description("isolate_host") == "Invoke 'isolate host' on this connector instance."
def test_capability_descriptions_handles_flat_strings() -> None:
"""Older connectors-service payloads emit ``["pull_alerts", ...]``."""
entry = {"capabilities": ["pull_alerts", "isolate_host"]}
out = _capability_descriptions(entry)
assert set(out.keys()) == {"pull_alerts", "isolate_host"}
# Defaults must look reasonable, not raw enum values.
assert "pull alerts" in out["pull_alerts"]
def test_capability_descriptions_handles_object_payload() -> None:
"""Forward-compat: connectors service may switch to richer objects."""
entry = {
"capabilities": [
{"value": "pull_alerts", "description": "Stream new detections."},
{"value": "isolate_host"}, # description omitted → fallback
],
}
out = _capability_descriptions(entry)
assert out["pull_alerts"] == "Stream new detections."
# Missing description falls back to the auto-generated one.
assert "isolate host" in out["isolate_host"]
def test_capability_descriptions_skips_unknown_shapes() -> None:
"""Garbage entries shouldn't crash — they should just be ignored."""
entry = {"capabilities": [123, None, ["nested"]]}
assert _capability_descriptions(entry) == {}
def test_capability_descriptions_handles_missing_capabilities_key() -> None:
"""Catalog entry without a ``capabilities`` key returns empty dict."""
assert _capability_descriptions({}) == {}
assert _capability_descriptions({"capabilities": None}) == {}