Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Monitor de performance do Claude Code e sistema local. Diagnostica lentidao, mede CPU/RAM/disco, verifica API latency e gera relatorios de saude do sistema.
.claude/skills/claude-monitor/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-09 | ✗→✓ | ▲ Improved | 18% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -25% | 0% |
| case-11 | ✗→✓ | ▲ Improved | 13% | 0% |
| case-16 | ✗→✓ | ▲ Improved | 46% | 0% |
| case-18 | ✗→✓ | ▲ Improved | 12% | 0% |
Monitor de performance do Claude Code e sistema local. Diagnostica lentidao, mede CPU/RAM/disco, verifica API latency e gera relatorios de saude do sistema.
Skill para diagnosticar e resolver problemas de lentidão no Claude Code e no sistema. Determina se o gargalo é local (PC) ou remoto (API Claude) e sugere ações corretivas.
Rode SEMPRE como primeiro passo:
bashpython C:\Users\renat\skills\claude-monitor\scripts\health_check.py
O script analisa em ~3 segundos:
O script retorna um JSON com diagnosis contendo:
bottleneck: "cpu" | "ram" | "browsers" | "disk" | "network" | "claude_api" | "ok"severity: "critical" | "warning" | "ok"suggestions: Lista de ações recomendadassummary: Resumo em português para mostrar ao usuárioMostre o summary ao usuário e ofereça executar as sugestões.
Baseado no diagnóstico, ofereça ao usuário:
bashpython C:\Users\renat\skills\claude-monitor\scripts\health_check.py --browsers-detail
Mostra RAM por browser e sugere quais fechar. Nunca fechar processos sem permissão explícita do usuário.
Se o usuário quiser monitoramento em background:
bashpython C:\Users\renat\skills\claude-monitor\scripts\monitor.py --interval 30 --duration 300
Parâmetros:
--interval: Segundos entre cada amostra (default: 30)--duration: Duração total em segundos (default: 300 = 5 min)--output: Caminho do arquivo de log (default: monitor_log.json)--alert-cpu: Threshold de CPU para alerta (default: 80)--alert-ram: Threshold de RAM % para alerta (default: 85)O monitor salva snapshots periódicos e gera um relatório ao final com:
Para testar se a lentidão é da API:
bashpython C:\Users\renat\skills\claude-monitor\scripts\api_bench.py
Mede o tempo de resposta do processo Claude Code local (não faz chamadas à API). Compara com tempos típicos e indica se está dentro do esperado.
| Métrica | OK | Warning | Critical | |---------|-----|---------|----------| | CPU % | <60% | 60-85% | >85% | | RAM usada % | <70% | 70-85% | >85% | | RAM browsers | <3 GB | 3-6 GB | >6 GB | | Processos browser | <30 | 30-60 | >60 | | Disco livre | >15% | 10-15% | <10% | | Latência rede | <200ms | 200-500ms | >500ms |
Quando apresentar o diagnóstico, inclua estas dicas contextuais:
50 abas = 50 processos competindo por recursos.
Mas se estiver usando >6 GB com várias sessões, considere fechar sessões antigas.
A sessão precisa carregar o histórico da conversa, e se o CPU está ocupado, demora.
lentidão generalizada.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 11,331 | 4,182 | -63% | 1 | 1 | 0% | 1,945 | 1,852 | -5% | 0 | 0 | — |
case-02 | fail→fail | 11,644 | 7,071 | -39% | 1 | 1 | 0% | 1,976 | 1,930 | -2% | 0 | 0 | — |
case-03 | fail→fail | 12,385 | 5,846 | -53% | 1 | 1 | 0% | 2,164 | 2,004 | -7% | 0 | 0 | — |
case-04 | fail→fail | 13,604 | 8,670 | -36% | 1 | 1 | 0% | 2,619 | 3,199 | +22% | 0 | 0 | — |
case-05 | fail→fail | 11,975 | 8,891 | -26% | 1 | 1 | 0% | 2,351 | 3,519 | +50% | 0 | 0 | — |
case-06 | fail→fail | 9,391 | 5,875 | -37% | 1 | 1 | 0% | 1,776 | 2,760 | +55% | 0 | 0 | — |
case-07 | fail→fail | 10,805 | 2,491 | -77% | 1 | 1 | 0% | 1,986 | 2,016 | +2% | 0 | 0 | — |
case-08 | fail→fail | 8,844 | 4,787 | -46% | 1 | 1 | 0% | 1,572 | 2,418 | +54% | 0 | 0 | — |
case-09 | fail→pass | 7,628 | 3,170 | -58% | 1 | 1 | 0% | 1,858 | 2,200 | +18% | 0 | 0 | — |
case-10 | fail→pass | 13,809 | 6,263 | -55% | 1 | 1 | 0% | 2,949 | 2,220 | -25% | 0 | 0 | — |
case-11 | fail→pass | 12,203 | 3,035 | -75% | 1 | 1 | 0% | 1,917 | 2,161 | +13% | 0 | 0 | — |
case-12 | fail→fail | 14,550 | 3,534 | -76% | 1 | 1 | 0% | 2,562 | 2,209 | -14% | 0 | 0 | — |
case-13 | fail→fail | 6,759 | 2,277 | -66% | 1 | 1 | 0% | 1,272 | 1,965 | +54% | 0 | 0 | — |
case-14 | fail→fail | 9,092 | 1,978 | -78% | 1 | 1 | 0% | 1,819 | 1,932 | +6% | 0 | 0 | — |
case-15 | pass→pass | 10,106 | 1,796 | -82% | 1 | 1 | 0% | 1,718 | 1,900 | +11% | 0 | 0 | — |
case-16 | fail→pass | 7,612 | 2,249 | -70% | 1 | 1 | 0% | 1,370 | 1,998 | +46% | 0 | 0 | — |
case-17 | pass→pass | 12,834 | 5,940 | -54% | 1 | 1 | 0% | 2,164 | 2,677 | +24% | 0 | 0 | — |
case-18 | fail→pass | 11,058 | 2,766 | -75% | 1 | 1 | 0% | 1,824 | 2,035 | +12% | 0 | 0 | — |
case-19 | fail→fail | 11,827 | 7,993 | -32% | 1 | 1 | 0% | 2,134 | 3,159 | +48% | 0 | 0 | — |
case-20 | pass→pass | 12,836 | 10,927 | -15% | 1 | 1 | 0% | 2,012 | 2,734 | +36% | 0 | 0 | — |
case-21 | fail→fail | 13,643 | 11,762 | -14% | 1 | 1 | 0% | 2,246 | 3,664 | +63% | 0 | 0 | — |
case-22 | fail→pass | 10,655 | 2,810 | -74% | 1 | 1 | 0% | 1,839 | 2,048 | +11% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 21 counted toward the lift figure. The other 1 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +27 percentage points is the difference between those two pass rates over the 21 comparable cases.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.