TODO.md

• sync up
• --watch flag for sync up. To keep the connection alive as long as --watch is running, without needing to persist the connection forever even once it stops running, in a separate thread have an empty shell open on the remote. (Or check if SSH supports a do-nothing command that doesn't open a shell, but keeps the connection active.)
• sync down
• .slink config directory, with an ignore file inside
• also allow a target file inside the directory-specific .slink config dir, to mirror a directory under a different path on the remote (or just to force a stable path in general)
• upload
• download
• forward ...
• slink go should accept an optional path. Mirror the local path if none is provided by default.
• Figure out how to safely canonicalize paths for download, where they don't exist on the local system but do exist on the remote
• Allow up, down, upload, and download to take an optional second argument to allow uploading/downloading/syncing to specific directories that don't match pwd on the remote machine
  • sync up
  • sync down
  • upload
  • download
• reset should pop back up to last configuration. Switched host configuration to use environment variables, so you don't need a reset: custom host configs are per-shell-session, and spawning a new shell sets you back to whatever your default exported env var is.
• clear should clear all host configuration and socket files
• current should print the current host
• Integration test slink by running an sshd in a Docker container
• Actually exit with correct exit codes rather than panicking
• Support kubectl as a transport. Upload and download can use kubectl cp instead of scp, go and run can use kubectl exec instead of ssh, and this ServerFault post explains how to use kubectl exec as an rsh-alike to get rsync to work. You'll need to alter how use works, to make it more like git remotes (since you need more configuration when you support multiple transports)... slink remote add <shortcut> ... and slink remote use <shortcut>? Allow setting a global default, but also allow per-directory transports and defaults in the .slink/ config directory.
• Allow nesting arbitrary transports via forwarding. Syntax TBD, but maybe something like --transport=forward(<transport> over(<shortcut> <port>)). Or better: slink remote add ssh <shorcut> ..., slink remote add kubectl <shortcut> ..., slink remote add forward <shortcut> --via <other-shortcut> --port <port> Which allows the type system to verify that flags for SSH aren't used for kubectl or forward, etc, via StructOpt. You can nest infinitely by making additional remotes that themselves are forwarded. This allows you to e.g. use a local kubectl to transport to a remote Kubernetes cluster that is inaccessible to the outside world by traversing through an SSH bastion.
• Allow per-command remote defaults in the slink config files, so that you can sync to one remote but e.g. run or go to another. This is useful if you're syncing files to a remote host over SSH, and the files are volume-mounted into containers run by a local Kubernetes cluster on the remote; you'd want the files to go to the underlying host, but go and run to jump into the containers.