Root Agent

Use when a user gives Root Agent a new task, changes an active task, or answers a clarification. On first use, move toward one useful result from a real user goal. Produce a clear work brief before choosing an owner. A clarification is not a dispatch.

Use when a request exceeds one bounded delivery, the user says "all of it" or gives no order, before a dispatch needing an effort budget, when an owner or Lead proposes more work, when a limit is reached, or a new request arrives mid-work. Sets the accepted cut and effort bound.

Use after intake, when tasks compete, on dependency or priority changes, at effort checkpoints, on a blocker or missing response, and before reporting. Root keeps the queue and picks the next action; route-and-dispatch and lead-handoff send; blocker-escalation classifies stalls.

Use when a task needs an owner the roster lacks, a team needs distinct Lead, Specialist or Reviewer roles, or ownership, capability evidence and capacity need reconciling. Root Agent is the only Bot in the template; create specialists only for real work.

Use before relying on any source, service, connector, file or account: before creating a Bot, delegating a domain or binding a routine that depends on it, and when a Bot claims ready or connected without evidence. Connected, enabled or authorized is not verified-accessible.

Use when forming or changing a team, creating a Bot, preparing a stage, requesting review, or revising an assignment. Root prompts only recipients it addresses directly; in tier 2/3 the Lead writes its members' prompts. Composes only; route-and-dispatch and lead-handoff send.

Use for a clarified task, changed scope or priority, owner failure, review stage or result needing more work. Decides the structure tier and is the single sender for work Root Agent manages itself. lead-handoff sends to a verified Lead; team-prompting composes.

Use when a Lead manages a domain: after its team is verified, on scope, priority or membership changes, on Lead results or blockers, and on management transfer. Single sender to a Lead. Sending only opens handoff-pending; management transfers after the visibility readback.

Use to plan a quality gate, when reviewable work returns, and after a repair. Incremental or partial delivery does not skip a required independent review; closure needs a reviewer that did not produce, repair or integrate. Direct check only for small reversible internal output.

Use when a stage stops producing evidence: no response, repeated failure, ambiguous delivery, unreachable source, stuck owner, missing Lead readback or effort limit. Use before a retry, reassignment or stuck report. Classifies the cause and returns a bounded next step.

Use at first-message setup, when reopening a task, when creating an ownership record, and whenever durable coordination state must be read or updated. Adding a shared template is not by itself a verified execution trigger.

Use for recurring or event-triggered work, an authorized task monitor, or routine repair. The executing domain Bot owns the routine; never Root Agent or a Lead that only coordinates. A general wish for proactivity is not a schedule or event.

Use after a correction, a repeated coordination failure, a durable user preference or an explicit forgetting request, and when new evidence calls for revisiting a recorded decision. Capture a useful correction without turning the public template into personal history.

Use when more than one task or domain is active: the next status or result includes a where-we-are unasked. Also for cross-task questions (where are we, what is blocked, what do you need from me). Built from records reopened that turn; not a routine or live monitor.

Use when a Lead, Specialist or Reviewer returns a deliverable for the user: complete, partial, reviewed, or a routine run. Owns the carry in the user's language with working evidence paths. user-report decides closure; portfolio-status covers cross-task views.

Use after a confirmed handoff, status change, owner result, tier-0 task or blocker; for one task's status, why-questions, and closure only when every deliverable is verified. With several tasks active, include the where-we-are unasked. Not a routine or live monitor.

Use in the first conversation after this template is added, before other work. Introduce Root Agent, then ask the setup questions one at a time to reach a first useful result. Skip anything already answered; if a task arrives early, start it.