Fleet Delegation Dispatch

skill

Translate exec asks into well-formed tasks for the right fleet agent and report back on completion.

Download .md

Fleet Delegation Dispatch

Turn an executive's ask into a well-formed task routed to the right fleet agent, then track it to completion and report back in plain language.

When to use

  • An executive hands you a request that another agent on the fleet is better suited to execute (research, outreach, data pull, follow-up).
  • You need to split a broad ask into concrete, assignable tasks with clear owners.
  • A delegated task has gone quiet and needs a status check or nudge.
  • Work is done and the executive needs a concise completion summary.

Tools

  • list-agents
  • create-task
  • delegate-task
  • check-task
  • send-message

Playbook

  1. Clarify the ask. Restate the executive's request in one or two sentences, including the outcome they actually want (not just the literal words), so the task you create is unambiguous.
  2. Call list-agents to see which agents are available and what each is scoped to do (for example, a research agent, an outreach agent, a scheduling agent). Match the ask to the agent whose scope fits best.
  3. Use create-task to write the task with a specific, checkable definition of done (for example, "Compile top 5 vendor candidates with pricing into a comparison list" rather than "look into vendors"). Include any deadline or urgency the executive gave you.
  4. Use delegate-task to assign the task to the chosen agent. If no agent's scope matches the ask, do not force it onto the closest one — say so and route the request back to a human via send-message explaining the capability gap.
  5. If the task involves booking travel or making any payment, do not delegate it for execution. Delegate only the research or proposal portion (options, prices, terms) and hand the actual booking or payment decision to a human via send-message.
  6. Poll progress with check-task at a reasonable cadence given the task's urgency. Do not check obsessively; check when the executive asks for an update or when the expected completion window has passed.
  7. If check-task shows the task stuck or blocked, use send-message to ask the assigned agent (or a human, if the blocker is outside any agent's authority) what is needed to unblock it.
  8. Once check-task shows the task complete, pull the result and translate it into a short, plain-language summary for the executive — what was found or done, and any decision or approval now needed from them.
  9. Send that summary to the executive via send-message, flagging clearly if a human step (approval, payment, booking) remains outstanding.

Failure modes

  • Delegating a vague task and getting back work that misses the executive's real intent — always restate the ask and confirm the definition of done before delegating.
  • Assigning a task to an agent whose scope does not cover it, producing unusable output — check list-agents scopes first, and escalate to a human when nothing fits.
  • Letting a task silently stall — check-task regularly and re-engage via send-message rather than assuming silence means progress.
  • Delegating a booking or payment step for execution instead of proposal-only — these must always end with a human decision.

Done when

  • The task has been delegated to the correctly scoped agent with a clear, checkable definition of done.
  • check-task confirms completion (or a documented blocker with an owner assigned to resolve it).
  • The executive has received a concise summary via send-message, with any required human action (approval, booking, payment) clearly called out.