openalice-world-agent-runner

Strong substrate reuse — wraps openalice-agent-basic AgentCore + agent-protocol-rs and mirrors the openalice-rt wire subset (like the TS rt-client); 0 circular deps, 0 dead imports, manifest 100, only 3 runtime deps — uniform with ecosystem conventions.

perceive→reason→act tick loop (TICK_MS) with clean code-health (0 bare excepts, 0 cycles), but no observed graceful-shutdown/SIGTERM, WS reconnect, retry, or timeout handling; max complexity 19 in a god module concentrates risk; E2E reliability is deploy-gated and unproven.

WS presence client (no server surface, so no rate-limit/headers/SSRF axis); handles RT_JWT_SECRET via env for room auth, but that secret plus RT_WS_URL sit in declared-unused env drift — auth wiring may be incomplete; no injection/authz signal to confirm, graded conservatively.

Manifest claims movement/presence logic is unit-tested, but no tests dir was detected in the clone and inspector reports no test count; live E2E against rt + Docker is explicitly deferred (deploy-gated) — unit-only at best, critical path untested.

Stateless by design — manifest owns zero tables, no migrations dir, no DB deps; correct for a headless presence runner, but there is no schema/constraint/RLS/GDPR evidence to credit, so scored low on absence.

No HTTP/health endpoint (health probe failed: 'no https endpoint in manifest exposes'), no metrics/tracing evidence, structured-logging unconfirmed (0 console statements but no logging signal) — a headless always-on runner with effectively no probes is a real gap.