Crate agent_trace_contract 

HTTP API contract for the Trace Infra service.

Why trace is its own service and not part of History Infra

Conversation messages and trace events look similar on the wire and behave nothing alike:

	Conversation	Trace
Replayed into a model prompt	yes	never
Losing one record	loses user work	acceptable
Volume	one per turn	many per turn
Retention	durable	short

Sharing one store would force the stricter guarantee onto high-volume diagnostic data, and — worse — put events one filtering mistake away from being replayed as conversation. Separate services keep that impossible by construction.

Layering

agent-trace-contract   paths + DTOs + redaction   (this crate)
agent-trace-runtime    store implementation
agent-trace-server     router + auth + bin

See api.md for the full HTTP reference.

Structs

ApiError

Error body shared by every non-2xx response.

AppendEventsRequest

AppendResponse

Result of an append. skipped counts ids that already existed, so a caller can tell a fresh write from a deduplicated retry.

EventsResponse

Events in chronological order, oldest first.

RuntimeEventRecord

One runtime trace event.

SearchEventsRequest

TraceScope

Identifies what a trace event belongs to.

Constants

AUDIENCE

DEFAULT_PAGE_SIZE

Default page size when a request omits limit.

DEFAULT_PORT

EVENTS_APPEND_PATH

EVENTS_SEARCH_PATH

HEALTH_PATH

MAX_PAGE_SIZE

Upper bound the server enforces on any page size.

REDACTED

Placeholder written over any value that looks like a credential.

SERVICE_NAME

SERVICE_TOKEN_ENV

Optional static service token. When unset, the server runs in open mode.

Functions

redact_payload

Mask credential-shaped fields anywhere in a trace payload.