Security

TL;DR

CMDOP uses a five-layer security model: TLS 1.3 encryption on all connections, OAuth 2.0 authentication with device code flow, workspace-based tenant isolation, session-level authorization with operator/observer roles, and process-level agent isolation with no privilege escalation. Agents connect outbound only, eliminating open ports as an attack surface.

CMDOP implements multiple layers of security to protect your machines and data.

What is the security architecture?

How does transport security work?

What TLS encryption is used?

All communication uses TLS 1.3:

Connection	Encryption	Certificate
Agent → Control Plane	TLS 1.3	Server cert validation
Client → Control Plane	TLS 1.3	Server cert validation
Control Plane → Centrifugo	TLS 1.3	Internal PKI

Can TLS be disabled?

There is no option to disable TLS in production. Development mode (--dev) uses a separate endpoint with self-signed certificates.

How does authentication work?

How does the OAuth 2.0 device flow work?

For CLI and agents, CMDOP uses OAuth Device Code flow:

How are tokens managed?

Token Type	Lifetime	Storage
Access Token	1 hour	Memory / secure file
Refresh Token	30 days	Encrypted file
API Key	Until revoked	User manages

How is token security enforced?

Tokens hashed (SHA-256) before storage in database
Automatic refresh 5 minutes before expiration
Revocation via API or dashboard
Last-used timestamp tracking

How does workspace isolation work?

How is multi-tenancy implemented?

What isolation guarantees are provided?

Resource	Isolation
Sessions	Workspace-scoped
Machines	Unique hostname per workspace
API Keys	Workspace-scoped
Audit Logs	Workspace-scoped

What workspace membership roles exist?

Owner: Full control, billing, member management
Admin: Machine management, user invites
Member: Access to sessions
Guest: Read-only access (optional)

How does session authorization work?

How is access control enforced?

Sessions enforce authorization at multiple levels:


# Three-level authorization check performed for every session access request
 
# Check 1: Verify user belongs to the workspace that owns this session
user.workspaces.contains(session.workspace)
 
# Check 2: Ensure session is active (not disconnected or pending)
session.status in [CONNECTED, BACKGROUND]
 
# Check 3: Verify user has an assigned role — OPERATOR can send input, OBSERVER is read-only
session.get_role(user) in [OPERATOR, OBSERVER]

What is the difference between Operator and Observer?

Permission	Operator	Observer
View output	✅	✅
Send input	✅	❌
Execute commands	✅	❌
File read	✅	✅
File write	✅	❌
Resize terminal	✅	❌

How is the agent secured?

Why is the agent outbound-only?

Agents never accept incoming connections:


# Traditional SSH: server listens on port 22, requiring inbound firewall rules
Traditional:
  Internet ──▶ [Firewall] ──▶ Server:22
  ❌ Attack surface: open port
  ❌ Requires firewall rules

# CMDOP: agent initiates outbound HTTPS — no listening ports, no firewall changes
CMDOP:
  Agent ──▶ [Outbound HTTPS] ──▶ Control Plane
  ✅ No open ports
  ✅ No firewall changes
  ✅ Works through corporate proxies

How does process isolation work?

Agent runs as the user who installed it
No root/admin privileges required
Commands execute with user’s permissions
No privilege escalation mechanism

What PTY security measures are in place?

Terminal emulation includes security measures:

Protection	Description
ANSI filtering	Blocks dangerous escape sequences
Input validation	Sanitizes control characters
Output limits	Prevents memory exhaustion

How does audit logging work?

What events are logged?

Event	Data Captured
Session created	User, machine, timestamp
Command executed	Command, exit code, duration
File accessed	Path, operation, timestamp
Authentication	User, method, IP, success/failure

How long are logs retained?

Audit logs retained for 90 days (configurable)
Logs stored separately from session data
Export available via API

What compliance standards are supported?

Audit logs support:

SOC 2 compliance
GDPR data access tracking
Internal security reviews

How should the API be secured?

What are the API key best practices?


# Recommended: set API key as an environment variable (not visible in source code)
export CMDOP_API_KEY="cmd_live_xxx"
cmdop connect
 
# Recommended: restrict config file permissions to owner-only read/write
chmod 600 ~/.cmdop/config.yaml
 
# Avoid: hardcoding keys in scripts exposes them in version control
# api_key = "cmd_live_xxx"  # Don't do this!

What rate limits are applied?

Endpoint	Limit
Authentication	10 req/min
Session operations	100 req/min
File operations	50 req/min

How should self-hosted deployments be secured?

For self-hosted deployments:

What network security is recommended?

What configuration is recommended for each component?

Component	Security Setting
PostgreSQL	TLS required, strong passwords
Redis	TLS, AUTH, no public exposure
Django	SECRET_KEY rotation, CSRF protection
Centrifugo	Token auth, channel permissions

What is the security checklist?

What should users do?

What should administrators do?

How do I report security issues?

Found a security vulnerability? Please report to: [email protected]

We follow responsible disclosure practices and appreciate your help keeping CMDOP secure.

Workspaces — Multi-tenant isolation
API Authentication — API key management

Last updated on April 17, 2026

Overview