You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Fix: <command> --help and <command> -h work for every command. Before, control --help printed {"error":"Unknown option --help.","ok":false} and exited 1, and control -h started the room. Now the help of a command shows its usage lines (from the top-level help), its options and, for control and attach, its keys, and exits 0. The command does not run.
New: help <command> shows the same text. help, --help and -h show the top-level help, which now has a help line. help <name> for a name that is not a command is an error (exit 1).
Unchanged: an unknown option still fails with Unknown option --<name>. (exit 1). A --help or -h that is the value of an option (send <session> --message --help) is text, not a help request.
Tests: tests/help.rs has one test per command (--help, -h and help <command>: exit 0, the same text, the command name in it), the options and keys of control, unknown options, and a check that each option in the usage has a help text.
Security: through MCP, remote_approve and remote_deny now do nothing by default. They return the exact command for the user to run (remote-sessions approve <session> --request <id>, /remote-approve, or a / d in control). A model cannot change the environment of the MCP server that already runs, so it cannot answer a permission prompt by itself. To let the MCP tools act, set REMOTE_SESSIONS_MCP_APPROVE=1 in the environment that starts Claude Code. The CLI approve and deny, /remote-approve and the control room are unchanged. Tests: tests/mcp_tools.rs (off for unset, 0, true, yes and empty, with no token read; on only with 1).
Security: the login token goes only to https://api.anthropic.com. REMOTE_SESSIONS_API now accepts only that address, or plain http to a loopback test server; before, any https address was accepted. Test: the_token_goes_only_to_api_anthropic_com (remote-sessions-core, src/api.rs).
New: several machines in the control room and in list. The Remote Control panel lists this machine (always, labeled with its short host name) and the machines of --host a,b (comma list, repeatable), REMOTE_SESSIONS_HOSTS and ~/.config/remote-sessions/hosts (one ssh host per line, # comments), merged in that order, each once. Each machine is listed with the list --host call (ssh -n -o BatchMode=yes -o ConnectTimeout=5), one thread per machine, stopped after 8 s; a machine that fails gets its own row (vm-1 ssh timed out) and never hides the others. Rows get a host column, and the panel title counts the sessions of each machine. Drop-in, approve, deny, read and say use the session API by claude.ai id, so they work for any machine; stop still needs --host (a signal over ssh). list (CLI and MCP remote_list) takes the same sources and gives each worker a host (one --host keeps the old output). /remote-room passes REMOTE_SESSIONS_HOSTS into a tmux split.
Security: an ssh host name is an ssh config name or a host name (letters, digits, ., _, -; user@ allowed); a leading - (-oProxyCommand=...), spaces and other characters are refused before ssh runs. Before, : and [] were also accepted. Tests: tests/hosts.rs (a stub ssh on PATH with two machines that answer and one that times out; the three sources and their order; the room rows, host column and counts; the refusals), rc.rs (host_names_refuse_options_and_odd_characters, a_slow_command_is_stopped_at_the_limit), sessions-tui/tests/room.rs (several_hosts_get_a_host_column_counts_and_a_row_for_a_host_that_failed).
Docs: README "Security and trust": your login, the hosts in the code, the risks, and how to report a security problem. README.zh-CN has a short version.
Versions: remote-sessions-cli, remote-sessions-core and sessions-tui are 0.3.1; the plugin is 0.3.1.