restart-serverlisted

# /restart-server — MCP + Kart Hot Reload Reload Willow, the Kart worker, and Grove. Willow hot-swaps a module whitelist in-process; tool bodies in `sap_mcp.py` and most `core.*` need **generation-swap** reload (`target="code"`) or a full process restart. The **Kart worker is a separate systemd service** (`kart-worker.service`) — merged Kart code stays stale until that unit is bounced, which `fleet_reload`/`fleet_restart` now do for you (idle-only). ## Operator opt-in: true hot reload (no MCP reconnect) Set `WILLOW_TRUE_HOTRELOAD=1` in **your** `.cursor/mcp.json` (or private MCP env). Do **not** commit this to the tracked public-fallback template unless the fleet opts in fleet-wide (ADR-20260704). With the flag on: - `fleet_reload(target="code")` — generation-swap tool bodies in-process - `fleet_reload(target="all")` — whitelist + Kart bounce, then **auto-chains** generation-swap when `code_version.stale` is still true ## TOOL PRE-LOAD (first action after invocation) ``` ToolSearch query: "select:fleet_reload,fleet_restart,fleet_system_status" ``` ## Sequence 1. **Hot reload Willow + Kart** — call `mcp__willow__fleet_reload` with target `"all"`. This hot-swaps the Willow module whitelist **and** bounces the `kart-worker` unit (skipped automatically if a Kart task is in-flight — check the `kart` field in the result and report it). With `WILLOW_TRUE_HOTRELOAD=1`, a still-stale `code_version` triggers generation-swap automatically. 2. **Read `code_version`