SentinelLAN supports authorized endpoint management. The current source is maintained on main; no long term support release is promised. This policy reflects the October 2026 source.

Use GitHub's private vulnerability reporting feature for this repository. Do not open a public issue with exploit details, credentials, private keys, or personal data. Include the affected commit, trust boundary, a minimal reproduction with dummy data, and expected/observed behavior. Maintainers coordinate the fix and disclosure.

For an exposed deployment, revoke affected credentials and follow the incident workflow. Never test a system without authorization.

• Human roles are Admin and Employee; Agent authentication is separate. Migration 20261009065038_RetireLegacyRoles locks unsupported roles (including Technician and PlatformOwner), changes security stamps, and revokes refresh sessions and activation tokens without granting replacement privileges.
• Production commands require RSA-PSS/SHA-256 with at least 3072-bit keys, versioned R2 envelopes and a key ID. Only the server holds SENTINELLAN_COMMAND_PRIVATE_KEY_PEM; Agents receive SENTINELLAN_COMMAND_PUBLIC_KEY and the matching key ID. Command HMAC fallback is limited to Development without RSA configuration. Agent HTTP requests instead use dedicated per-device HMAC keys, a ±60-second timestamp, a persisted one-use nonce and an exact body hash; the original secret bootstrap closes after the first signed proof. See ADR 0029.
• Keep secrets outside Git in environment variables/user secrets; separate command signing, access-token signing, vault and data-encryption keys. Sensitive support data uses authenticated encryption at rest. Attachment lists select metadata and a persisted plaintext byte count; only an authorized download loads image content.
• API policies, use cases and EF global tenant filters enforce tenant scope and Employee ownership/assignment. Filters do not replace authorization or provide PostgreSQL row-level security.
• Optional host monitoring reads a bounded local JSON snapshot from a read-only mount. A root-owned systemd collector produces it. The API has no SSH credentials or Docker socket and exposes no VPS restart/reboot operations. Only the configured operating organization's Admin can read host status; see host monitoring.
• Forwarded headers run before authentication/rate limiting. Configure SENTINELLAN_KNOWN_PROXIES and/or bounded SENTINELLAN_KNOWN_NETWORKS for the immediate proxy. Defaults trust loopback only. Unknown peers cannot supply client IP/scheme; one hop is consumed and paired header chains must have equal length. Do not expose API/container ports directly.
• Commands require allow-list, tenant/device checks, reason, confirmation, expiry, nonce, signature and audit. SimulateLock/SimulateNetworkIsolation always simulate; real Windows/Linux lock/isolation additionally requires explicit lab/device authorization. See execution limits.
• Collect technical telemetry only. Support images are employee-selected and consented. Audit is append-only through the application, with no independent WORM guarantee.

See the threat model for operational limits and validation requirements.

Redis backplane uses a deployment-specific channel prefix; keep it private and authenticated. Rate limiting remains per process and needs ingress limits when scaling. S3/MinIO attachment objects use AES-GCM with tenant/request/attachment AAD, and authorized downloads remain behind the API. Back up the bucket, database and encryption keys together. Production ASP.NET runtime and self-contained Agent releases require 10.0.13 or later. Deployment details and limits.