Docs / Examples
Agent operations example
An operations agent can use hosted capabilities without receiving unrestricted access to production systems.
Plain English
The agent asks hosts to do narrow things, and each host enforces lifecycle, permission, and evidence.
Why it exists
Operational agents need useful actions, but infrastructure teams need accountable boundaries and replayable decisions.
Formal definition
Agent operations uses CHP capabilities as governed action surfaces around ticket classification, status updates, and incident notifications.
Concrete example
Ground the concept before the schema.
Ops Agent invokes classify_ticket and request_incident_update with audited policy and evidence export.
Classifies an operational ticket and proposes a routing label.
Invocation trace
- 01 Actor
Ops Agent
- 02 Capability
classify_ticket
- 03 Host
OpsSupportHost
- 04 Policy
audited
- 05 Context
ticket_context
- 06 Result
Routing Label
{ "invocation_id": "inv_session_abc_001", "capability_id": "schedule_technician", "version": "1.0.0", "mode": "sync", "correlation": { "correlation_id": "session-abc" }, "subject": { "id": "agent://planning-assistant", "roles": ["dispatcher"] }, "payload": { "job_id": "job_456", "window": "tomorrow" }, "requested_at": "2026-06-16T15:14:20.000Z"}Relationships
Where this sits in the protocol.
Each concept should explain its neighbors so implementation teams can preserve the boundary across manifests, invocation, evidence, and tests.
Agent frameworks discover host capabilities.
Applications govern high-value operations.
Infrastructure providers export evidence to telemetry systems.
Visual model
- 01Agent selects a capability.
- 02Host checks policy and context.
- 03Outcome and evidence are visible to operators.
Implementation notes
- Keep dangerous actions blocked or approval_required by default.
- Use audited policy for low-risk read and classify actions.
- Treat unavailable hosts as protocol outcomes.
Common mistakes
- Giving the agent broad credentials instead of host-scoped capabilities.
- Skipping evidence for read-like actions.
- Treating host unavailability as agent failure.
Related concepts