Harness Context

Retrieve the neutral session-bootstrap context for a service so your own framework can bootstrap against the same world model the hosted runner uses.

The harness context endpoint returns the neutral, session-bootstrap context for a given service. This is the same projection the platform's hosted runner renders from - identity, reference instructions, world scope, tool descriptors, guardrails, and the server-enforced write floor - so a customer's own framework can bootstrap a session against the same world model.

The projection is PHI-free: no scoped entities or rendered caller prose are included.

Endpoint

Method	Path	Auth	Rate Limit
GET	/v1/{workspace_id}/agent-runs/harness-context	Workspace API key or operator identity token	Read

Query Parameters

Parameter	Type	Required	Default	Description
service_id	UUID	Yes	-	The service whose context to retrieve.
version_set	string	No	release	Which version set to resolve the service configuration from (e.g., release, draft). Max 255 characters.

Response

Returns the full harness context object for the resolved service configuration.

Field	Type	Description
context_version	integer	Schema version of the context payload.
config_fingerprint	string	Fingerprint of the resolved configuration.
source	string	How the context was produced (e.g., api).
identity	object or null	Agent identity and reference instructions. null for native runs where the customer's own definition carries identity.
world_scope	object	World-model read/subject scope: workspace_id, scoped_entity_ids, allow_create.
tools	object	Neutral tool descriptors available to the agent: descriptors, read_tool_names, write_tool_names. Discovery hint only - enforcement happens at the world-tools edge.
guardrails	object	Guardrail configuration: guardrails, global_action_guidelines, global_boundary_constraints.
write_floor	object	The server-enforced write bound: allow_create, requires_entity_anchor, enrichment_write_scope, clinical_write_scope, clinical_write_principal, unanchored_or_api_key_writes, enforced_at. Advisory to render - the server enforces it regardless of what a client does.
runtime	object	Runtime hints: engage_model, language, channel_kind, caller_context.

Identity Object

Present only for platform-authored services (not native runs).

Field	Type	Description
entry_actor_name	string	Display name of the entry actor.
name	string	Agent name.
role	string	Agent role description.
developed_by	string	Developer attribution.
instructions	string	Reference instructions for the agent.
initial_state	string	Initial conversation state.

Example Request

curl -X GET \
  'https://{base_url}/v1/{workspace_id}/agent-runs/harness-context?service_id={service_id}&version_set=release' \
  -H 'Authorization: Bearer {api_key}'

Example Response

{
  "context_version": 1,
  "config_fingerprint": "a1b2c3d4",
  "source": "api",
  "identity": {
    "entry_actor_name": "Front Desk",
    "name": "Alex",
    "role": "Patient coordinator",
    "developed_by": "Acme Health",
    "instructions": "Help patients coordinate appointments...",
    "initial_state": "greeting"
  },
  "world_scope": { ... },
  "tools": { ... },
  "guardrails": { ... },
  "write_floor": { ... },
  "runtime": { ... }
}

Notes

• The context is byte-identical to what the hosted runner renders, so remote and hosted sessions bootstrap from the same world model.
• The endpoint resolves the service configuration through the same path used by hosted runs, including version-set handling and 404 behavior for missing services.
• PHI-free: no scoped entities or rendered caller prose are included in the response.
• Read-rate-limited under the same policy as other read endpoints.