On-Call Scheduling

SkillMonitoring & ops

Read and change ServiceNow on-call rotations — the cmn_rota → cmn_rota_roster → cmn_rota_member chain, how "who is on call right now" is resolved from a member window plus the schedule's time zone, and how escalation from an incident reaches it.

Available today. Use it from your connected AI after setup.

Connect ahel once, and every AI you use reads what you have installed.

Then ask your AI: use the On-Call Scheduling skill

What this skill tells your AI

The instructions your AI receives, as published by serac-labs/serac in packages/skills/oncall-scheduling/SKILL.md and read by ahel’s review.

"Who is on call for the database team?" is one question and four tables. Answering it by guessing at table names is how agents end up querying sys_user_group.manager and reporting the wrong person.

The chain

sys_user_group          the team being covered
   └── cmn_rota          a rotation on that group ("Database — primary")
        ├── cmn_schedule the hours the rotation covers, and its TIME ZONE
        └── cmn_rota_roster   a rotation pattern: weekly, 7 days, starting <date>
             └── cmn_rota_member   one person, with a from/to window
                  └── sys_user

Read it as: a group has rotations, a rotation has rosters, a roster has members. The roster is the pattern (rotate weekly, one week each). The member row is one person's turn — it carries from and to as glide_date, and member referencing sys_user.

A group can have several rotations at once — primary, secondary, escalation — and they are unrelated records. "The on-call person" is never a property of the group.

Who is on call right now

snow_oncall_manage({ action: "get_current_oncall", rotation_sys_id }) resolves it in three steps, and knowing them tells you what a wrong answer means:

  1. Read cmn_rota_roster where rota=<rotation>^active=true.
  2. Find the cmn_rota_member rows belonging to those rosters whose from/to window covers today.
  3. Resolve member to a sys_user.

Two consequences worth having in front of you:

  • A rotation with no member window covering today returns nothing. That is not an error; it is a rotation whose roster has run past its generated windows. The tool says so explicitly rather than guessing at the next person.
  • The dates are dates, and coverage is hours. cmn_rota_member.from/.to are days. Whether the person is on call at 03:00 is decided by the rotation's cmn_schedule — including its time zone, which is a field on the schedule and not on the rotation. A rotation whose schedule is in US/Pacific hands over at a different UTC moment than the same rotation in Europe/Amsterdam. Never convert on-call times yourself without reading cmn_schedule.time_zone first.

Full ServiceNow resolution goes through the OnCallRotation script include, which also applies escalation order and time-off records. The tool's answer is the coverage row; if an instance has time-off overrides in play, cross-check with that script include via snow_execute_script before paging anyone.

Listing and swapping

// Rotations for a group, by name or sys_id
await snow_oncall_manage({ action: "list_rotations", assignment_group: "Database" })

// Shifts in a window. Defaults to now → now + 7 days.
// Times are ServiceNow datetime format, UTC: "YYYY-MM-DD HH:MM:SS"
await snow_oncall_manage({ action: "list_shifts", rotation_sys_id: rota, end_time: "2026-09-01 00:00:00" })

// A one-off swap: reassign a member row from one person to another
await snow_oncall_manage({
  action: "swap_shift",
  roster_sys_id: memberRowSysId,     // a cmn_rota_member sys_id, despite the parameter name
  from_member_sys_id: currentUser,   // guard: refuses if the row does not hold this user
  to_member_sys_id: replacement,
})

Two traps in that last call:

  • roster_sys_id takes a cmn_rota_member sys_id — the per-window assignment row, not a cmn_rota_roster sys_id. Take it from list_shifts, do not go looking for a roster.
  • reason is accepted and not persisted. cmn_rota_member has no reason or override column; the tool echoes it back in the response and nothing on the instance records why the swap happened. If an audit trail matters, write a work note on whatever record prompted the swap.

A swap edits one window. It does not change the pattern — next rotation the original person is back on.

Schedules underneath

cmn_schedule is shared machinery: SLAs, business-hours calculations and on-call rotations all point at it. Build it before the thing that uses it.

const sched = await snow_create_schedule({ name: "Follow the sun — EMEA", type: "weekly", time_zone: "Europe/Amsterdam" })

// Spans include or exclude time on top of the base schedule
await snow_add_schedule_entry({ schedule_sys_id: sched.sys_id, name: "Kings Day", type: "exclude",
                                start_date_time: "2026-04-27 00:00:00", end_date_time: "2026-04-27 23:59:59" })

type: "exclude" is how holidays and blackout windows are modelled — as a span that removes time, not by editing the weekly pattern. include adds coverage the base pattern does not have. A schedule with no spans covers nothing.

Changing a schedule changes every SLA that references it. Check before editing:

await snow_query_table({ table: "contract_sla", query: "schedule=" + sched.sys_id, fields: "name,active" })

Why anyone asks: escalation from an incident

The usual path is: a P1 lands on a group, nobody picks it up, and something has to page a human. That "something" reads the group's rotation, not the group.

// 1. the incident's assignment group
const inc = await snow_query_table({ table: "incident", query: "number=INC0012345",
                                     fields: "assignment_group,priority,state" })
// 2. its rotations
const rotas = await snow_oncall_manage({ action: "list_rotations",
                                         assignment_group: inc.records[0].assignment_group.value })
// 3. who is covering the primary one right now
const who = await snow_oncall_manage({ action: "get_current_oncall", rotation_sys_id: rotas.rotations[0].sys_id })

If step 2 comes back empty, the group has no rotation and there is no on-call person to find — the honest answer is "this group is not on call", not the group's manager.

Related

  • sla-management — the other consumer of cmn_schedule, for business-hours SLA calculation.
  • incident-management — what escalation does once it has the name.

Signals

GitHub stars
78
Forks
26
Last commit
Sep 2026
Advanced
Catalog kind
skill
Gateway key
oncall-scheduling
Source
github.com/serac-labs/serac