Codeman 1.31.0: your own model server in the Run menu, and waking a sleeping box

[Ark0N]

Codeman 1.31.0 is out. Two new features, and a batch of terminal fixes that came out of people using the thing and noticing when it lied to them.

Point a session at your own model server. #393 landed the backend for custom model endpoints and left it HTTP-API-only. It is now the whole feature: turn on Custom model endpoints in App Settings, save an endpoint, and the Run dropdown grows a Custom Endpoints section built live off the CLI registry, one entry per harness that can actually redirect plus each endpoint you saved. Pick one and it launches that harness pointed at your server, asking which model first when the endpoint has more than one. Endpoints re-discover themselves every five minutes, and one unreachable endpoint never blocks the others. Thanks @opticon454, who validated this against a real llama-swap box rather than against the tests, and tracked both the /props versus /running context discrepancy and the DeepSeek /v1 failure down to the SDK source instead of guessing at them.

Wake a sleeping remote host. If you run a remote SSH case against a machine that suspends, the session was there and the host was not, and typing into it went nowhere. Now a host can carry a wake target, either a MAC address for Wake-on-LAN or a wake command of your own, and Codeman uses it when you ask for that host: when you type into a sleeping session, when you press wake on the banner, or when you start or attach a session on it. Anything you type while it wakes is held and flushed once it is back. Waking only ever happens because you asked for that host, never off a watcher or a reconnect, since a machine woken by a reconnect watcher just comes back seconds after every suspend. Thanks @Randalix, who built this against a real sleeping machine and then re-read the whole diff between rounds rather than only the parts I asked about.

Prompts sent through the API actually get submitted now. Claude Code 2.1.277 started ignoring Enter for the first 30 to 50 seconds after the composer paints, while still accepting the typed text. So a prompt sent right after a session came up sat there unsent, and every waiter (send-and-wait, the agent skill, cron, my own maintainer bot) burned its full timeout on a turn that never started. The server now re-presses Enter on a schedule, but only while the composer verifiably still holds the text it sent. If the composer is empty, or holds something else, or there is no composer at all, it stops.

The Instance count stepper works on every run mode. It only ever applied to Claude. Set it to 3, press Run on OpenCode or Codex or Grok, and you got exactly one session, with no error and nothing to tell you the control had done nothing. All eight of the other run modes honour it now, and the banner says how many are starting. Thanks @timkjr, and that is a first contribution.

Three terminal fixes from @irisitymichaelgrundberg, who has been finding the cases where what you see is not what the pane actually holds:

• Opening a session could draw a frame built for a pane bigger than your terminal, which quietly ate rows. On a 50-row pane in a 30-row terminal, 28 of a 45-line command survived, drawn twice. The response now reports the geometry a capture was taken at, so the browser can see the mismatch and fix it.
• Anything the CLI printed between a capture and the end of loading it used to be dropped, so the next partial redraw landed on a frame your terminal had never seen. Most visible as garbled output right after a tab switch.
• Copying out of a pane put a wall of spaces on the clipboard. A full-screen program paints real spaces across the rest of every row, and those count as content: 138 of them per line in a 282-column pane. Copy trims them now.

On that last one, the PR also stripped the leading indent every selected row shares, and I dropped that half before merging. It fires on 73% of real non-TUI text, and no width threshold separates a program's margin from content that is genuinely indented, so it was quietly rewriting git log bodies, indented Python and git diff context lines. The trailing trim is the uncontroversial part and that is what shipped.

Full notes: https://github.com/Ark0N/Codeman/releases/tag/codeman%401.31.0

Thanks

@opticon454, @irisitymichaelgrundberg, @timkjr and @Randalix.