# Delegated tasks, messages and artifacts Canonical: https://semiaifoundry.com/learn/atlas/delegated-agent-protocols/ Revision: 2026-10-07 Rights: https://semiaifoundry.com/ai-atlas/licenses.html Delegated tasks and coordination · Atlas v1.3.0 # Delegated tasks, messages and artifacts Task lifecycle, artifact ownership, input-required states, cancellation and authenticated service boundaries. [Explore in Atlas](https://semiaifoundry.com/ai-atlas/#delegated-agent-protocols)Logical state, task evidence and declared units; no physical assembly scale Learning objectiveTrack task identity, partial artifacts and input-required states without equating remote completion with accepted task success. ## The mechanism Delegation hands a bounded task to another participant and receives messages, artifacts and status over time. That interaction can outlive a single request. The parent must know which task is being updated, who owns its artifacts and whether it is working, waiting for input, completed, failed or canceled. A2A defines a task-oriented protocol surface for such exchanges; this does not prove that either participant is intelligent or correct. A tool call and a delegated task can both be asynchronous, but they may expose different lifecycles and ownership contracts. The delegator should not turn a dropped connection into a new task without checking the existing identity. It should also retain its own final-goal and authority checks rather than treating a remote completed status as permission to publish an artifact. 1. The request declares a task identity, goal and permitted information/effects for the receiving service. 2. Status and artifact updates refer to that identity while execution continues. 3. An input-required state asks for missing information without claiming completion or starting a duplicate task. 4. The parent verifies the returned artifact and applies its own task and permission contract before further effects. ## Science and assumptions ### Delegated tasks, messages and artifacts · explicit relationship A delegated task retains its identity across messages, input requests and artifact updates. Remote completion is checked against the parent’s task and authority before further effects. A fictional delegated review task T17 returns a partial inspection note while its state is working. It later requests the asset revision and remains input-required. Sending that revision resumes T17; creating unrelated T18 instead could duplicate work and lose the original context. When T17 reports completion with an artifact, the parent checks that it refers to FAN-7 and the requested revision. A completed artifact for FAN-9 remains wrong for this task. This is a worked message lifecycle, not a running A2A endpoint. Assumptions Worked lifecycle based on the inspected A2A specification; no network endpoint, interoperability test or multi-agent runtime is executed. The concrete protocol revision must be pinned for a future adapter. ## Engineering example A fictional delegated review task T17 returns a partial inspection note while its state is working. It later requests the asset revision and remains input-required. Sending that revision resumes T17; creating unrelated T18 instead could duplicate work and lose the original context. When T17 reports completion with an artifact, the parent checks that it refers to FAN-7 and the requested revision. A completed artifact for FAN-9 remains wrong for this task. This is a worked message lifecycle, not a running A2A endpoint. ## Distinctions and tradeoffs Delegation does not transfer unlimited authority or make a remote completion flag an independent correctness verdict. Task-oriented protocols support long-lived work but require explicit ownership, reconnection and failure handling. Coordinated AI systems need identities and outcome contracts between participants, not merely more agent names. ## Specifications and interfaces Mechanism boundary How is an unfinished delegated task represented? Worked lifecycle based on the inspected A2A specification; no network endpoint, interoperability test or multi-agent runtime is executed. The concrete protocol revision must be pinned for a future adapter. Evidence type Authored worked example Worked explanation with a bounded numeric or state example. This lesson does not claim an executing implementation of an external model or platform. ### Interfaces Declared input and state Input The request declares a task identity, goal and permitted information/effects for the receiving service. Checked outcome Output Track task identity, partial artifacts and input-required states without equating remote completion with accepted task success. ## Check your understanding ### The network disconnects while delegated task T17 is working. What should the parent avoid? 1. Checking T17’s existing status 2. Blindly creating a duplicate task for the same intended effect 3. Retaining the task identity Answer and reasoning Blindly creating a duplicate task for the same intended effect A lost connection does not establish that the task stopped; duplication must be controlled at the task/effect boundary. ## Evidence and scope This is a worked explanation, not an executing external platform, model or robot. The quantities and fixtures are authored. Source scope is recorded in the automation learning review manifest; no source benchmark or private system behavior is claimed. Worked lifecycle based on the inspected A2A specification; no network endpoint, interoperability test or multi-agent runtime is executed. The concrete protocol revision must be pinned for a future adapter. The worked example explains this mechanism. General numerical labs are a reference; they do not execute this lesson’s full mechanism. ## Sources - [A2A specification](https://a2a-protocol.org/latest/specification/) - [Temporal Activities](https://docs.temporal.io/activities) [Published lesson data](https://semiaifoundry.com/ai-atlas/lesson-data/delegated-agent-protocols-f7d8f254309a16f8.json) · [Atlas releases and licensing](https://semiaifoundry.com/ai-atlas/licenses.html) ## Continue exploring ### Foundations - [A tool protocol is a boundary](https://semiaifoundry.com/learn/atlas/mcp-host-client-server/) - [Durable actions and the ambiguous timeout](https://semiaifoundry.com/learn/atlas/durable-actions-idempotency/) ### Related lessons - [Tools, memory & agent loop](https://semiaifoundry.com/learn/atlas/tool-loop/) - [Evaluation & the AGI frontier](https://semiaifoundry.com/learn/atlas/evaluation/)