@staff0rd/assist 0.413.0 → 0.414.0

package/README.md

@@ -305,6 +305,8 @@ From WSL, the selector can also surface and drive Windows-host repos (requires `
 - `sessions.windowsDaemonHost` / `sessions.windowsDaemonPort` — where the WSL daemon reaches the native Windows daemon (defaults `127.0.0.1` / `51764`; set the host to the Windows IP on WSL2 NAT-mode networking).
 - `sessions.windowsVersionCheck` — how the WSL↔Windows daemon handshake reacts to a protocol-version mismatch: `block` (default) refuses creates and auto-heals the host, `warn` logs and proceeds anyway, `off` skips the check. Use `warn`/`off` to keep working across an unfixable version gap.
 
+A `run:` entry in `assist.yml` flagged `server: true` (with an optional display-only `port:`) is treated as a singleton **dev server**: at most one may be live per normalised git remote (`getCurrentOrigin`), i.e. across a clone and all its sibling clones of the same repo. Every session card for a repo that defines a `server: true` run shows a **▶ start** button (surfaced by `GET /api/server-runs?cwd=`); clicking it sends `create-run`. The daemon enforces the singleton on `create-run` (and on relaunch via retry) — if a server run is already live for that remote it rejects with a conflict instead of starting a second one that would collide on the port. The web UI turns that conflict into a "replace running server?" prompt; confirming stops the old server and starts the new one, declining leaves it untouched. The serving session's own card shows a `serving :<port>` chip and a **⏹ stop** button (stop leaves a relaunchable card rather than dismissing it), and the slot frees automatically whenever that session stops, exits, or is killed. Non-`server` runs (`build`, `test`, …) are unconstrained.
+
 When iterating on assist itself: web server changes only need the `assist sessions` process restarted — sessions survive. Daemon/session-core changes need `assist daemon restart` to load the new code; this kills the PTYs, then claude sessions — including assist sessions that wrap claude, like `assist draft` — are auto-respawned via `claude --resume` with scrollback starting fresh, while run sessions (and assist sessions whose claude sessionId was never discovered) reappear as not-restored tiles that can be retried. A `--once` draft/bug/refine session is respawned through its assist wrapper instead of bare `claude --resume`, so its `assist signal done` watcher is re-established and the card still auto-closes.
 
 - `assist next [id] [--once]` - Alias for `backlog next [id]`; `--once` exits after the first completed item run instead of prompting for another