Upstream bug report

Target: github.com/openclaw/openclaw/issues

Title

4.26 doctor surfaces memory-readiness "not checked" warning on every run; previous behavior (4.25) only warned on actual probe timeout

Environment

• OpenClaw 2026.4.26 (be8c246)
• Memory provider: ollama (configured)
• Install: pnpm-detected, real path under .npm-global/lib/node_modules/openclaw

Summary

Per CHANGELOG line 181 in 4.26: "Memory/doctor: treat the specific gateway timeout after ... gateway memory probe result as inconclusive instead of reporting embeddings not ready, while preserving warnings for explicit failures." This is a positive change — it stops false alarms when the deep probe times out.

However, the new behavior also causes the doctor to surface a warning on every run, even when there's no probe timeout to be inconclusive about. The exact text:

Memory search provider "ollama" is configured, but the gateway reports embeddings are not ready. Gateway memory probe for default agent is not ready: memory embedding readiness not checked; run openclaw memory status --deep to probe Verify: openclaw memory status --deep

Concern

The warning fires on a non-deterministic state ("not checked") rather than an actual failure. For a healthy memory subsystem, asking the operator to manually run --deep on every doctor run is noise — doctor should either:

• Run the deep probe automatically and report the result, OR
• Skip the warning entirely when the configured provider is reachable, OR
• Surface this only when there's an actionable issue (e.g. failed connectivity)

Source location (4.26)

• dist/server-methods-b3jaTRE_.js:3764-3772 — skipped probe returns "memory embedding readiness not checked"

Reproduction

1. On 4.26, configure memory.search.provider: "ollama" (or any provider)
2. Run openclaw doctor
3. Observe the memory-readiness warning regardless of actual memory health

Production impact

LOW — purely cosmetic / advisory noise in doctor output. But it adds a known-noise item to the post-update doctor delta, making it harder to spot real new warnings against a baseline.

Workaround / ground-truth verified

Run openclaw memory status --deep separately to verify true memory health. We confirmed the warning is benign on a healthy install:

openclaw memory status --deep
# exit code 0
# Provider: ollama
# Model: nomic-embed-text
# Indexed: 740/793 files · 4361 chunks
# Embeddings: ready
# Vector: ready
# FTS: ready

Add this --deep invocation to post-flight smoke procedures so operators can verify the warning is doctor noise vs an actual issue.

Suggested fix

Either:

• Doctor should run the lightweight readiness probe automatically (don't ask operator to do it)
• OR: change the warning to a debug-level log (not a doctor-surfaced WARN)
• OR: surface only when probe explicitly fails or times out

The current state ("not checked, please check") is the worst of both worlds.