Summary

Add RTK support for ssh commands to filter remote output before it enters context.

Use case

When running remote diagnostics via SSH (e.g. ssh user@host "systemctl status ..."), raw output can be large. RTK already handles local equivalents (rtk ps, rtk df, rtk docker) — the same filtering should apply when those run over SSH.

Example

Current (unfiltered)

ssh name@xxx.xxx.x.xxx "systemctl status ollama && ollama ps && nvidia-smi"

Desired

rtk ssh ssh name@xxx.xxx.x.xxx "systemctl status ollama && ollama ps && nvidia-smi"

Privacy/security requirement

Output must be filtered LOCALLY — no credentials, hostnames, IPs, or command output leave the local machine. RTK acts as a local stdout filter on the subprocess return, identical to how it handles local commands today. No telemetry on remote host, command, or content.

Behavior spec

• rtk ssh spawns ssh as a subprocess (same as rtk ps spawns ps)
• Stdout filtered/compressed locally before entering context
• Falls back to raw output if filtering fails
• Respects existing rtk output size limits

Context

Discovered via rtk discover — 4 SSH calls in one session producing unfiltered remote service/GPU status. Local equivalents filter at 84–96% reduction; same savings expected.