Published
VSSH: Guarded SSH for Scripts and AI Agents
VSSH runs remote commands through native OpenSSH and adds what scripts and AI agents need: preflight guardrails, one-object JSON results, one-step file transfer and a privacy-safe audit trail.
Impact
- Open source on npm as
@light-merlin-dark/vssh, MIT licensed, released through npm trusted publishing with SLSA provenance. Documentation at vssh.io - Rebuilt as VSSH 2 in July 2026 around the native
sshandscpexecutables; the VSSH 1 MCP server, plugin runtime and credential store were removed to shrink the security surface - In daily use as the remote-operations path for the author's AI agents on production infrastructure
- Install in one line:
npm install -g @light-merlin-dark/vssh, thenvssh --setupandvssh doctor
Key Features
- Native transport: the same OpenSSH authentication, host verification,
known_hostsand ssh-agent behavior, with no VSSH-specific remote language - Preflight guardrails: recognizable root deletion, raw disk writes, destructive volume cleanup, firewall flushes and shutdowns are blocked before connecting
- Structured results:
--jsonreturns one bounded object with stdout, stderr, exit code, duration and timeout state - Transfer with permissions:
upload --mode 600copies a file and sets its mode as one reported operation - Diagnosis:
vssh doctorchecks binaries, configuration, identity and a real connection - Privacy-safe audit: owner-only outcome metadata and a command hash, never command text, output or credentials
- Agent discovery:
commands --jsonexposes a versioned, machine-readable command contract
Continue reading