Summary

openclaw doctor reports that local embeddings are “not confirmed ready” even when openclaw memory status --deep confirms the full memory stack is healthy.

Version

OpenClaw 2026.6.6 (8c802aa)

Reproduction

1. Configure agents.defaults.memorySearch.provider as local using the bundled Hugging Face GGUF embedding model.
2. Run openclaw memory status --deep.
3. Observe Embeddings: ready, Vector store: ready, Semantic vectors: ready, and FTS: ready.
4. Run openclaw doctor or openclaw doctor --deep.
5. Doctor still reports:

Memory search provider is set to "local", but local embeddings are not confirmed ready.
Gateway probe: memory embedding readiness not checked; run `openclaw memory status --deep` to probe

openclaw doctor --lint --json simultaneously returns {"ok":true,"checksRun":22,"checksSkipped":0,"findings":[]}.

Cause observed in installed build

doctor-B9vJA5aF.js defines SKIPPED_MEMORY_EMBEDDING_PROBE with checked: false. The doctor flow supplies that result to doctor-memory-search-C3qfCGTd.js, whose local-provider branch emits the warning unless gatewayMemoryProbe.checked && gatewayMemoryProbe.ready.

The --deep CLI option currently does not cause this memory probe to be checked, so it produces the same false warning.

Expected behavior

Doctor should either:

• run the memory embedding readiness probe (especially with --deep) and suppress the warning when ready, or
• report that readiness was not tested as neutral/informational rather than implying the configured local provider is broken.

Actual health

The authoritative deep memory command passes all components, so this appears to be a doctor false positive rather than a memory runtime failure.