Module observability

Modules§

dora
log
multi
noop
prometheus
runtime_trace: Compatibility shim for the doctor command’s log-reading utilities.
traits
verbose

AgentTurnGuard: Drop-safe lifecycle bracket for one logical agent turn.
BroadcastHookGuard
FlushGuard: Guard that flushes its observer on drop — the telemetry analogue of AgentTurnGuard. Held for the lifetime of a short-lived agent invocation (today: the CLI one-shot, zeroclaw agent -m ...), whose process exits before the OTLP batch exporter / metric PeriodicReader’s background interval fires. Without this flush all buffered telemetry — including the never-ended gen_ai.agent.invoke span, which is only .end()’d inside Observer::flush — is lost when the runtime is torn down.
LogObserver: Log-based observer — uses tracing, zero external deps
MultiObserver: Combine multiple observers — fan-out events to all backends
NoopObserver: Zero-overhead observer — all methods compile to nothing
PrometheusObserver: Prometheus-backed observer — exposes metrics for scraping via /metrics.
VerboseObserver: Human-readable progress observer for interactive CLI sessions. This observer prints compact > / < progress lines without exposing prompt contents. It is intended to be opt-in (e.g. --verbose).

ObserverEvent: Discrete events emitted by the agent runtime for observability.

clear_broadcast_hook: Remove the broadcast hook, if any. Intended for tests and orderly shutdown.
create_observer: Factory: create the right observer from config
set_broadcast_hook: Install a process-wide observer that will receive every event recorded through observers built by create_observer. Calling this again replaces the previous hook.
set_scoped_broadcast_hook: Install a process-wide observer and return a guard that clears it on drop.