Example Canonical Task Records

The schema in action

Three example tasks pulled from §3 / Appendix B, with their canonical fields populated. Shows the schema in §7.2 in action — and how much is inferred vs explicit at creation time.

D.1 Client booking request (assist tier)

A TA gets a message: "We need a 4-night stay at the Aman Tokyo for our anniversary, last week of October — suite with city view."

Field Value How populated
id tsk_01HXY... System-generated
title Search: Aman Tokyo, suite, anniversary last week of Oct Auto-generated from message content
action Search hotels matching property + dates + suite type; return options to TA for review Inferred from message intent
actor TA (with AI search assistance) Captured at creation
when immediately Default for Client-driven Assist
primary_owner TA on the trip Routing-assigned from trip ownership
watchers None
status In progress Lifecycle
primary_context trip:<Anderson family Tokyo Oct 2026> Inferred from conversation thread
origin Client-driven (explicit) From §3.1
creator AI message classifier Captured at creation
creation_surface Message bubble — inline suggestion Captured at creation
routing_tier Assist Default for Client-driven from §6.4
priority Normal Default; no urgency markers in message
due_at None set
rationale Client message requested hotel booking with specific property + dates + room type. AI
evidence_refs [link to message] AI
approval_state human-approved TA accepted the AI-proposed task
confidence_score 0.92 High — message specific, no ambiguity
task_type client_follow_up Inferred
subtype direct_booking_request Inferred
created_at 2026-05-12T09:43Z System
updated_at 2026-05-12T09:44Z System
snooze_untilNone set
dismissal_reasonNone set
reopen_count0System default
last_reopened_atNone
parent_task_idTop-level task (not a sub-task)
agent_run_idNot from agent mode

D.2 Weather-disruption alert (escalate tier)

Flight-tracking feed pushes an alert: typhoon warning for Seoul, client arriving in 48 hrs.

Field Value How populated
id tsk_01HXZ... System
title Typhoon warning Seoul — client arrival in 48hr Auto-generated from signal
action Alert client of typhoon warning + prepare reroute scenarios Inferred from signal + active trip
actor TA on the trip (with AI for reroute option generation) Routing-assigned
when immediately Time-sensitive disruption
primary_owner TA on the trip Routing-assigned
watchers Ops manager Auto-added (escalate-tier policy)
status In progress Lifecycle
primary_context trip:<Wei Seoul May 2026> Inferred from active-trip query
origin World-driven (disruption) From §3.2
creator AI signal monitor Captured at creation
creation_surface Weather feed monitor Captured at creation
routing_tier Escalate Default for World-driven (disruption) from §6.4
priority High Auto-bumped by time-to-departure proximity
due_at 2026-05-14T06:00Z (T-24hr from arrival) Derived from trip arrival timestamp
rationale Typhoon warning issued for Seoul; client arriving 2026-05-14. Operator should alert client and prepare reroute scenarios. AI
evidence_refs [weather feed alert], [trip itinerary] AI
approval_state v1: pending-review (every suggestion goes to human review). v2: auto-approved for high-confidence disruptions See §7.1 Stage 3
confidence_score 0.97 High — reliable weather feed, real active trip, clear time pressure
task_type disruption_response Inferred
subtype weather_disruption Inferred
created_at 2026-05-12T11:02Z System
updated_at 2026-05-12T11:02Z System
snooze_untilNone set
dismissal_reasonNone set
reopen_count0System default
last_reopened_atNone
parent_task_idTop-level task (not a sub-task)
agent_run_idNot from agent mode

D.3 Pre-arrival checklist (auto tier)

Daily scheduled query finds a trip departing in 10 days.

Field Value How populated
id tsk_01HY0... System
title Pre-arrival prep: Anderson Tokyo trip T-10d Templated from trip name
action Re-confirm transfers; brief restaurants on dietary notes; send weather + packing notes to client Templated from pre-arrival policy
actor Ops associate (with AI for templated communications) Routing-assigned
when scheduled:T-10d-from-departure Derived from trip-lifecycle rule
primary_owner Ops associate on the trip Routing-assigned
watchers TA on the trip Auto-added per pre-arrival policy
status In progress Lifecycle
primary_context trip:<Anderson family Tokyo Oct 2026> Inferred from scheduled query result
origin Trip lifecycle (templated prep) From §3.3
creator AI scheduled query Captured
creation_surface Scheduled query (daily, trips departing in N days) Captured
routing_tier Auto Default for Trip lifecycle templated prep from §6.4
priority Normal Default; will auto-bump as date approaches
due_at 2026-10-15T00:00Z (T-3d from departure) Derived from trip departure
rationale Trip departing in 10 days — pre-arrival checklist auto-spawned. AI (templated)
evidence_refs [scheduled query run id], [trip record] AI
approval_state v1: pending-review (templated trip-prep tasks still go to human review at launch). v2: auto-approved for high-confidence templates See §7.1 Stage 3
confidence_score 0.99 High — pure templated query, clear active trip, well-known pattern
task_type trip_lifecycle_prep Inferred
subtype pre_arrival_checklist_T-10 Templated
created_at 2026-10-05T00:01Z System (when the daily query ran)
updated_at 2026-10-05T00:01Z System
snooze_untilNone set
dismissal_reasonNone set
reopen_count0System default
last_reopened_atNone
parent_task_idTop-level task (not a sub-task)
agent_run_idNot from agent mode

Reading the examples

  • D.1 (Client-driven, Assist) — AI proposes, human approves; routing into the TA's task list. Most fields auto-filled from the message.
  • D.2 (World-driven disruption, Escalate) — AI auto-creates the suggestion; the routing tier bumps watchers and priority. Human gate behaviour is phased — v1 always routes to human review for fast triage; v2 auto-approves above the confidence threshold.
  • D.3 (Trip lifecycle prep, Auto) — Pure templated. The scheduled query knows it's a pre-arrival task and fills almost everything.

In all three: the operator never types 23 fields. The system or upstream signal does almost all the work. Manual creation is the only case where humans see fields directly — and per the §7.2 constraint, even then it's just title + owner.