OpenClaw merged PR #103476, "feat: correlate managed agent turns with ClawRouter requests", adding a cleaner diagnostics path between managed ClickClack agent turns, OpenClaw model-call diagnostics, and ClawRouter audit events.
This is an operator feature more than an end-user UI change. It gives teams a way to connect one inbound managed-agent event to the model request that handled it without widening logs to capture more content than necessary.
The Observability Gap
Before this patch, managed ClickClack agent turns could not be joined reliably to the exact OpenClaw diagnostic model call and ClawRouter audit event. Operators had timestamps and workload-level attribution, but not a strong shared identifier across the whole path.
That makes incident analysis harder. If several agent turns happen close together, timestamp matching can become ambiguous. The natural temptation is to log more data, including message content or transport-specific fields. That is exactly the pressure a privacy-conscious diagnostics system should avoid.
What OpenClaw Now Carries
The change carries ClickClack's validated payload correlation value across authoritative fetch and reply requests. It also derives a deterministic agent-mode run id from canonical message ids, then maps each diagnostic model-call id into ClawRouter's request-id contract.
Valid short ids stay exact. Opaque or overlong ids are converted into a deterministic ASCII-safe representation capped at 128 characters while preserving the :model:<n> suffix. Explicit case-insensitive request headers still take precedence.
The PR is careful about what it does not add: no message content and no transport-specific core diagnostic schema.
Why That Matters
Good observability is not just about seeing more. In agent systems, it is often about seeing the right metadata and resisting the urge to collect everything else.
With this patch, an operator can follow a managed turn from the inbound ClickClack event through the OpenClaw agent run and model diagnostics into ClawRouter's metadata-only audit trail. That is useful for debugging routing, attribution, provider behavior, and incident evidence.
It also keeps existing ClawRouter client, agent, session, credential, and explicit-header behavior unchanged. In other words, the correlation layer is additive rather than a rewrite of the routing model.
Validation
PR #103476 reports exact-head CI with 44 jobs passed, plus CodeQL, critical quality, workflow sanity, OpenGrep, and iOS dead-code workflows passing at the same head.
Focused coverage included 68 tests across diagnostic, ClawRouter, ClickClack, and managed-gateway end-to-end shards. The managed-gateway E2E path booted with an env SecretRef, proved readiness and catalog behavior, completed an attributed agent turn, and asserted ClawRouter request IDs stay within the 128-character contract while session attribution stays within 256 characters.
The PR also reports source-blind CLI/Gateway proof showing client=openclaw, agent=main, project/session attribution, a :model:<n> request-id suffix, explicit-header precedence, no credential observation, and no content captured in the report.
User Impact
For most users, this will show up indirectly as better supportability. When a managed agent turn needs investigation, operators have a cleaner way to connect the event, the OpenClaw run, and the ClawRouter request without guessing from time windows.
For teams running ClawRouter in managed-agent deployments, that means faster incident reconstruction and less pressure to increase log sensitivity just to make audits possible.
Bottom Line
PR #103476 gives OpenClaw a stronger metadata trail for managed ClickClack turns. It improves ClawRouter diagnostics by carrying deterministic correlation through the request path while explicitly avoiding message-content logging.