You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
This commit was created on GitHub.com and signed with GitHub’s verified signature.
Added
Huntress connector.huntress reads Huntress Managed EDR incident
reports — the footholds, ransomware canaries, process, antivirus, identity
and SIEM detections a Huntress SOC analyst has confirmed and delivered —
through GET /v1/incident_reports with HTTP Basic auth (#1845, part of #356). The API host is fixed, so the credential is only ever sent to
Huntress, and both halves of it are vaulted. The list has no time filter,
so it is read by updated_at, newest first, at the documented 500-report
maximum through read_token_pages, following pagination.next_page_token
(never next_page_url) until a page passes the start of the window. The
window is then applied to sent_at, closed at the moment the read began:
a report is raised once, when Huntress delivers it, and not again each
time it is closed, commented on or remediated; a report never delivered is
not raised. A report names its host only by agent id, so each distinct
agent is looked up once through GET /v1/agents/{id} for its hostname,
addresses and OS — at most 20 a poll and stopping on a 429, because the
account allows 60 requests a minute — and a failed lookup costs the host,
never the report. Severity keeps Huntress's low, high and critical,
and critical stays critical. Setup and troubleshooting are on the new Huntress page.
Security
An empty password can neither be set nor used (GHSA-xh3r-vqgc-cv78). POST /api/v1/tenants/me/users declared its password as a bare string, so
an explicit "" passed validation, was bcrypt-hashed like any password,
and POST /api/v1/auth/login accepted the empty string for that account
from then on — verify_password("", bcrypt("")) is simply true. Creation
now enforces NIST SP 800-63B's eight-character floor (and refuses past
bcrypt's 72-byte truncation boundary rather than silently cutting), and
login refuses the empty string at the boundary, so an account row created
with an empty password before the floor landed is locked out rather
than quietly serviceable forever — an admin recovers it by setting a real
password. SCIM and invite-pending accounts are unaffected (their sentinel
markers are non-empty and already fail verification). Reported by
abdochtouki with a live PoC.
Fixed
UEBA, honeytokens and purple-team tracing no longer points at the
service's own container (#1848). All three built an OTLP span exporter
inline in main.py against settings.otel_endpoint, whose default of http://localhost:4317 is — inside a container — the service itself, so
tracing could never reach the collector (otel-collector:4317, behind the
compose monitoring profile) and UEBA's batch processor retried a refused
connection every few seconds, hundreds of export-failure log lines on a
stand without the profile. The three now share a guarded app/core/telemetry.py (held byte-identical by scripts/sync_vendored_telemetry.py --check in CI): quiet by default
(OTEL_EXPORTER=none — deliberately diverging from api/agents' otlp
default, because these services emit spans continuously and a missing
collector must cost one log line, not a retry loop), with OTEL_EXPORTER=otlp targeting OTEL_EXPORTER_OTLP_ENDPOINT
(default http://otel-collector:4317, never localhost). Compose passes
both knobs through. The dead otel_endpoint config field is removed —
a deployment that set OTEL_ENDPOINT/UEBA_OTEL_ENDPOINT should set OTEL_EXPORTER=otlp and OTEL_EXPORTER_OTLP_ENDPOINT instead. Reported
by @43n51c with the repro and the fix sketch this follows.