openclaw: AbortSignal.timeout(5000) too aggressive on cold gateway init

[kaghni]

Summary

After PR #97 (feat/centralize-autoupdate), the OpenClaw plugin's checkForUpdate() correctly polls the npm registry instead of ClawHub, and emits Run: hivemind update instead of openclaw plugins update hivemind. But during cold gateway init, the 5-second AbortSignal.timeout(5000) on the registry fetch fires before the response lands, swallowing the upgrade-available notice for that gateway run.

Symptom

[plugins] Hivemind plugin registered
[plugins] Auto-update check failed: The operation was aborted due to timeout

Expected output ⬆️ Hivemind update available: <current> → <latest>. Run: hivemind update never renders.

Repro

1. Start a fresh openclaw gateway with --dev profile (or production)
2. Watch /tmp/openclaw.log for the [plugins] lines

The timeout is reproducible on cold start. Re-running the gateway tends to succeed (warm TLS / DNS).

Confirmation that the change itself is in the bundle

$ grep -oE 'registry.npmjs.org/@deeplake|clawhub.ai' ~/.openclaw/extensions/hivemind/dist/index.js
registry.npmjs.org/@deeplake

$ grep -c 'openclaw plugins update' ~/.openclaw/extensions/hivemind/dist/index.js
0

So the centralization at the source-of-truth layer (npm registry vs ClawHub) and the advice-text layer (Run: hivemind update) is in place. Only the live notice render is unreliable due to the cold-init timeout.

Confirmation the registry itself is fast

$ time curl -sS -o /dev/null https://registry.npmjs.org/@deeplake/hivemind/latest
real    0m0.167s

So 5s is plenty in steady state. The first-run timeout was likely cold TLS + bonjour-watchdog noise running concurrently with plugin register.

Suggested fix

In openclaw/src/index.ts:checkForUpdate:

1. Bump the timeout from 5000 to 10000 ms to absorb cold-init latency.
2. OR: on AbortError, retry once with exponential backoff before silently giving up.

const res = await fetch(VERSION_URL, { signal: AbortSignal.timeout(10000) });  // was 5000

What this PR established for OpenClaw

• ✅ Bundle URL changed from clawhub.ai → registry.npmjs.org/@deeplake/...
• ✅ Advice text changed from openclaw plugins update hivemind → Run: hivemind update (in the in-prompt notice, the registerCommand handler, AND the inlined SKILL.md)
• ⚠️ Live notice render not observed in E2E this run (5s timeout); registry measured 167ms in steady state, so a small bump fixes it.

Priority

Low. The check runs once per gateway start and doesn't affect runtime behavior. Users still get hivemind update (manually or through any other agent's session-start centralized trigger) — just don't get the in-prompt nudge from OpenClaw on the cold-start gateway run.