Update Kanboard

Kanboard.md

@@ -11,6 +11,13 @@ comments and moves cards like a regular team member via the dedicated
 - Port: `127.0.0.1:4800` → `:80` (reverse-proxied via Caddy externally)
 - Storage: SQLite in the `kanboard_data` volume (`/var/www/app/data/db.sqlite`)
 
+## Tier-0: backups, monitoring, degraded mode
+
+Kanboard is **tier-0** (fabric backbone, DESIGN-a2a-agents.md v2.1 §6c, hardened
+in kb#158/A2A-26) — scheduled backups, live Zabbix monitoring of the service and
+JSON-RPC API, and a verified degraded mode (Kanboard down → Adolf still chats,
+workers park and retry, nothing is lost). Full detail: [[Backups#Kanboard]].
+
 ## API
 
 JSON-RPC at `/jsonrpc.php`. Two auth modes:
@@ -45,6 +52,11 @@ registered in `~/.claude.json` as server **`kanboard`** (tools appear as
 (read straight from `mcp/.env`); writes are authored as `claude` via `user_id`.
 Env: `PORT`, `KANBOARD_URL`, `KANBOARD_TOKEN`, `KANBOARD_BOT_USER=claude`.
 
+Every tool handler wraps its call in try/catch and returns an MCP `isError: true`
+result on failure rather than crashing (verified kb#158) — so when Kanboard is
+unreachable, a `kanboard_*` tool call just fails cleanly; Adolf/Claude keep
+working normally otherwise.
+
 ```bash
 cd /home/alvis/kanboard/mcp && docker compose up -d --build   # (with proxy vars unset)
 curl -s http://localhost:3103/health                          # {"status":"ok","tools":23}