SKILL NAME inbound-ticket-triage-router WHAT IT AUTOMATES Incoming support tickets arrive faster than a human can read, categorize, and route them, so urgent issues sit behind trivial ones. This skill reads each new ticket, assigns a category, severity, and owning queue, and keeps working through the backlog until the unrouted queue is empty — turning a growing pile into a continuously drained stream. WHY A LOOP The backlog changes every cycle: new tickets land while the agent works, and earlier routing decisions surface patterns (a spike of the same bug) that should change how later tickets are handled. Each pass observes the current queue state and adjusts, which a single call cannot do. LOOP DESIGN Trigger: a new ticket lands in the unrouted queue, or the queue is non-empty on a scheduled sweep. Perceive: the oldest unrouted ticket's text, sender history, and the current counts per category/queue. Decide: classify category and severity from the ticket text; if a category is spiking this session, flag it for a possible incident rather than routing normally. Act: apply category and severity tags and move the ticket to the owning queue via the helpdesk API. Observe: confirm the ticket left the unrouted queue and the target queue count incremented. Exit condition: the unrouted queue is empty, OR a max of N tickets processed this run, OR a per-run time budget is reached — whichever comes first. BUILD NOTES Tools/data: helpdesk API with read + tag + move scopes (assumption — verify the specific platform, e.g. Zendesk/Intercom, exposes queue-move); ticket history lookup (assumption). Failure mode: misclassifying a high-severity ticket as low, burying it — guard by routing anything matching severity keywords or a low-confidence classification to a human review queue instead of auto-closing. Human approval: none needed for tagging/moving, but auto-escalating a cluster to a formal incident should require human confirmation before it pages anyone.