OpenClaw System Performance Diagnostics and Optimization

Problem

In OpenClaw-style LLM gateway systems, performance issues often appear as slow responses, CPU or memory usage above 80%, context usage above 70%, cache hit rates below 50%, or API timeouts. These problems may not come from a single failure; they can combine long conversation history, poor cache settings, dependency jitter, scheduled job peaks, or abnormal token cost.

How the skill works

system-diagnostics organizes debugging into four steps: quick diagnosis, deep analysis, immediate remediation, and long-term optimization. It checks a fixed set of thresholds: CPU, memory, context, cache hit rate, API success rate, and response time. Each symptom maps to concrete actions: for high CPU, inspect the gateway service, clean temporary files, and review concurrent connections; for high memory, trace leaks and review cache strategy; for high context usage, clean session history and tune pruning parameters; for low cache hit rate, warm frequently used entries and adjust TTL; for API failures, validate network, keys, timeout, and retry behavior.

Boundary

The skill is best suited to operational environments that already have OpenClaw gateways, session management, caching, and API call paths. It focuses on diagnostics and configuration tuning rather than replacing a full APM platform, capacity planning system, or self-healing automation. Users should feed real logs, metrics, and service configuration into the analysis; if baseline monitoring is missing, establish metric collection first, otherwise threshold checks remain mostly heuristic.