Support Triage

Tickets arrive triaged, not stacked

Every incoming ticket is classified, scored, and checked against what you already know — before the first person looks at the queue.

Use this solution

Sits behind your helpdeskTickets in via the data APIReplies stay human-sent

Support tickets triaged in a Subterranean workspace

Triaged before you look

Tickets arrive as rows and get classified, scored for severity, and matched against known issues before anyone opens the queue.

The second report is cheap

Every resolution either matches a known issue or writes one, so the same problem never costs a full investigation twice.

Nothing sends itself

The agent that runs unattended holds no tool that can reach a customer. Replies are drafted for a person to send.

The team

One agent triages, one agent resolves

The split is the safety model: the agent that wakes on its own can read, classify, and flag, but it has no way to talk to a customer.

S
Support EngineerSupport Interface
Does the work a person asked for — reproduce, diagnose, draft the reply, file the bug — and records what was learned on the way out.
T
Triage OfficerAutonomous Ticket Triage
Never talks to a customer. Wakes on each new ticket to classify it, set severity from observed blast radius, and check it against the known-issue ledger.

VisualA ticket landing and being classified against known issues

Included template

See exactly what gets deployed

Every agent, instruction, skill, and starter file is open to inspect before you create a workspace.

Agents: 2
Skills: 7
Files: 5

Inspect the complete template

/SUBTERRANEAN.md

Open file

# Workspace

This workspace is a support desk. It sits behind your helpdesk rather than
replacing it: tickets arrive here as rows, agents triage and investigate them,
and the answers go back out through the helpdesk a human already uses.

- /support/SETUP.md — table schema and the one-time setup checklist.
- /support/CONVENTIONS.md — the contract every agent follows when writing
  records. Read it before writing, not after.
- /support/INTAKE.md — how tickets get in. Rewritten during setup with the real
  table ids and a working request.
- /support/DIAGNOSTICS.md — what to ask a customer for, by ticket type.

## The operating rule

Ticket #1 costs an hour of investigation. Tickets #2 through #200 should cost

Ready to stop re-investigating the same bug?

Start from the template and have a triaged queue on the first day of tickets.

Use this solution Inspect the template