Turn "what's about to breach?" into a four-tier board. Remaining time sets the base tier; the
thread moves tickets up or down.
1. Establish the XLA scheme: per-priority response and resolution targets as the user states
them, or the priority mapping from available levels with the desk's defaults. If no targets
are known, ask — never invent one.
2. Pull open tickets, split per board and per priority. Sweep Honesty base skill: disclose caps
as "at least N" — a missed board here is exactly the ticket that breaches.
3. For each ticket compute remaining time to the nearest applicable target — first response for
unanswered tickets, resolution otherwise — respecting pauses where a waiting status
legitimately stops the XLA clock, but verify the wait is real (a client-visible ask exists),
not just a parked status. State the math in business hours with the assumption shown, and say
when the calendar is unknown.
4. Assign the base tier by remaining time: Breached (target passed), Critical (breaches this
shift, default under 4 business hours), High (breaches within the next business day), Watch
(within the tier horizon, default 3 business days).
5. Adjust one tier up or down for factors read from the thread, citing the factor that moved it —
no vibes-based bumps. Escalating: client frustration or executive contact, repeat issue,
unassigned ticket, a prior breach on the ticket, contractual or VIP flags the user defined.
Mitigating: fix delivered pending client confirmation, a new timeline agreed in writing, an
appointment before the target. Breached never adjusts down — a breach is a fact.
6. Recommend one move per ticket: assign now, escalate to a named role, respond now
(first-response saves are cheapest), request a pause with documented client agreement, or
accept-and-communicate an unavoidable breach.
7. Output the board: Breached first with age past target, then Critical, High and Watch, each
sorted by remaining time. Per line: ticket, client, priority, remaining or overdue time,
adjustment factor, recommended move. On request, leave a plain-text risk-flag note or assign
and escalate — with confirmation, never in bulk.
If the scheme is unknown or a ticket's target is ambiguous, put it in an UNKNOWN bucket and say
what's missing. Report breaches without spin: age past target, not "slightly over".
Run unattended from an external scheduler: the entire reply is a plain-text tier digest, no
narration. Board list, per-priority targets, tier horizons and business calendar are supplied,
never inferred; without targets reply exactly "XLA TARGETS NOT CONFIGURED - NO REPORT." Ambiguous
targets go to UNKNOWN. Step 5 does NOT run there — thread-evidence bumps stay attended, so tier
by remaining time only. No writes. Zero open tickets in scope, reply exactly "NO TICKETS AT
RISK."
Magic Library