Action domain

Autonomous & Agent Actions

Records what agents and automated systems did at execution time—tool calls, delegated actions, approvals, and state mutations—without becoming a monitoring system and without interpreting correctness, intent, or harm.

Receipt moment

Issue a receipt at the tool boundary: when an agent invokes an external capability or commits an action.

issueReceipt({ actor: "agent", action: "invoke_tool|execute_task|approve_step|modify_resource", object: "tool|task|resource", ref: "run-id-or-task-id", policy: "agent-policy-version" // optional })

Preserved fields

agent identity (abstracted) · action · tool/resource reference · timestamp · stable run reference
Optional: guardrail/policy version, delegation reference, correlation reference

Common integration surfaces

Agent runtimes, orchestration frameworks, automation platforms, decision engines, and robotics control loops.

Why this avoids “source of truth” liability

When autonomous behavior is questioned later, providers are pressured to narrate what happened. DAR enables a safer posture: provide verifiable receipts of the actions executed at time T, and keep interpretation outside the platform.

Example events

agent_tool_invoked email_sent_on_behalf_of_user ticket_status_changed api_key_rotated_by_agent payment_refund_initiated_by_agent

Back to hub

Action Domains → #agents