OpenClaw ElonTools Optimizer

Otimizador completo de performance, retenção de contexto e economia de tokens para OpenClaw. Configurações battle-tested de produção (ElonTools) que resolvem...

Security Scan

The skill's files and runtime instructions are consistent with an OpenClaw configuration optimizer: presets and a config.patch workflow match the stated purpose, but review before applying because it can change agent/plugin configuration and includes troubleshooting commands that touch host files.

Purpose & Capability

Name and description claim an OpenClaw optimizer; the included presets, settings guide, and examples (read preset then gateway(action="config.patch", raw=...)) all directly implement that purpose. No unrelated binaries or credentials are requested.

Instruction Scope

Most instructions stay within scope (reading packaged presets and applying them via gateway config.patch). However the Troubleshooting section contains an explicit shell command to remove lock files (rm /data/.openclaw/agents/*/sessions/*.lock) and references host paths (e.g., memory/YYYY-MM-DD.md), which are out-of-package actions and could modify host state if executed. The SKILL.md also includes guidance to disable/enable plugins globally via config.patch — expected for the optimizer but impactful.

Install Mechanism

Instruction-only skill with no install spec and no bundled executable code; nothing is written to disk by the skill itself beyond the packaged preset files. Lowest install risk.

Credentials

The skill declares no required environment variables, credentials, or config paths. It does not request secrets. It does reference system paths and gateway configuration APIs (expected for a config patch tool), but does not require elevated credentials in metadata.

Persistence & Privilege

The skill does not request always:true and has no install, but its intended use is to patch global agent configuration (disabling plugins, changing bootstrap/system prompt limits, session/prune behavior). That is coherent for an optimizer but means it can change other components' behavior; apply with caution and prefer testing on staging first.

Scan Findings in Context

[system-prompt-override] expected: A pattern flagged as 'system-prompt-override' was detected. This skill intentionally adjusts bootstrap/system-prompt-related settings (bootstrapMaxChars) which can be considered a form of system-prompt modification — that's expected for a prompt/optimizer. Still, because system-prompt changes can alter agent behavior, review the exact config values before applying.

Assessment

This skill appears coherent for tuning OpenClaw performance, but do not apply presets blindly on production. Recommended steps before installing/applying: 1) Backup current OpenClaw gateway configuration. 2) Inspect the included preset JSON and the settings-guide to confirm they match your needs (especially plugin entries that are disabled). 3) Test on a staging instance first; confirm no needed channel plugins are disabled. 4) Do not run host-modifying troubleshooting commands (like the rm /data/... lockfile line) unless you understand the impact and have backups. 5) Note the skill has no homepage and an unknown source — prefer installing from a trusted source or mirror that you can audit. If you want higher assurance, ask the author for provenance or run the config.patch operation in a dry-run or with explicit diffs before accepting changes.

OpenClaw + ElonTools Optimizer v2

Configurações de produção testadas em ambiente real (ElonTools — 9 agentes, 14+ deploys/dia, sessões de 5h+).

O Que Mudou na v2

Novidade	Impacto
Heartbeat com Haiku	~10x mais barato que Opus por heartbeat
Disable plugins ociosos	Menos memória, startup mais rápido
web_fetch maxChars cap	Evita páginas de 200K chars no contexto
Bootstrap maxChars	Controla tamanho do system prompt
Sub-agent auto-archive	Limpa sessões de sub-agentes automaticamente

Aplicação Rápida

Ler o preset desejado e aplicar via gateway config.patch:

# Ler o preset
read("references/preset-balanced.json")

# Aplicar
gateway(action="config.patch", raw=<conteúdo do preset>)

⚠️ Nota sobre Plugins

Os presets desabilitam plugins de canais sem configuração (WhatsApp, Discord, Slack, Nostr, Google Chat, iMessage, Signal). Se você usa algum desses canais, remova a entrada correspondente do preset antes de aplicar, ou re-habilite depois:

// Exemplo: re-habilitar Discord
gateway(action="config.patch", raw='{"plugins":{"entries":{"discord":{"enabled":true}}}}')

Presets

Preset	Arquivo	Uso
balanced	references/preset-balanced.json	✅ Recomendado — equilíbrio entre retenção e performance
max-retention	references/preset-max-retention.json	Sessões longas de código — máxima retenção de contexto
lightweight	references/preset-lightweight.json	Máquinas fracas — overhead mínimo

O Que Cada Preset Configura

Configuração	Lightweight	Balanced	Max Retention
Context Pruning TTL	15min	30min	45min
Soft Trim (contexto %)	50%	60%	70%
Hard Clear (contexto %)	75%	85%	90%
Soft Trim max chars	4K	8K	12K
Reserve Tokens	50K	40K	30K
History Share pós-compact	50%	70%	80%
Memory Flush trigger	80K tokens	120K tokens	140K tokens
Session Prune	3 dias	7 dias	14 dias
Max Entries/session	200	500	1000
Timeout por turno	5 min	15 min	15 min
Bootstrap max chars	10K	15K	20K
web_fetch max chars	15K	30K	50K
Heartbeat model	Haiku	Haiku	Haiku
Sub-agent archive	15min	30min	60min
Plugins ociosos	Disabled	Disabled	Disabled

Como Funciona (Fluxo Visual)

📊 Contexto crescendo normalmente...

[0-50%] ✅ Nada acontece — tudo normal

[60%] ⚡ SOFT TRIM ativa
  → Tool results > 30min e > 500 chars
  → Reduz para: [500 chars início] ... [500 chars fim]

[85%] 🔥 HARD CLEAR ativa
  → Tool results expirados removidos completamente
  → Placeholder: "[tool result removed to save context]"

[120K tokens] 💾 MEMORY FLUSH preventivo
  → Salva contexto automaticamente em memory/YYYY-MM-DD.md

[160K tokens] 🗜️ COMPACTAÇÃO
  → Memory flush final → Resume histórico → Mantém 70% recente

Resultado: Compactação ~2-3x menos frequente. Quando acontece, contexto crítico já está salvo em memory/.

Economia de Tokens (v2)

Otimização	Economia estimada
Heartbeat com Haiku	~10x por heartbeat (24 heartbeats/dia = muito!)
web_fetch cap 30K	Evita páginas de 100K+ chars enchendo contexto
Bootstrap 15K	Reduz system prompt (menos tokens/turno)
Plugins disabled	Startup mais rápido, menos overhead
Sub-agent archive 30min	Limpa sessões órfãs automaticamente

Detalhes dos Parâmetros

Ver references/settings-guide.md para explicação completa de cada parâmetro.

Troubleshooting

Problema	Solução
Contexto perdido após compactação	Aplicar preset balanced ou max-retention
Agente lento/travando	Aplicar preset lightweight
Lock files stale após crash	rm /data/.openclaw/agents/*/sessions/*.lock + restart
Sessões acumulando disco	Verificar session.maintenance.mode: "enforce"
Memory flush não criando arquivos	Verificar que memory/ existe no workspace
Plugin re-habilitado não funciona	Restart gateway após config.patch
web_fetch truncando demais	Aumentar tools.web.fetch.maxChars

Checklist Pós-Aplicação

1. ✅ gateway(action="config.get") — verificar config aplicada
2. ✅ Sem warnings na resposta
3. ✅ Verificar que plugins que você USA estão enabled
4. ✅ Monitorar próxima compactação — mais contexto retido
5. ✅ Verificar memory/YYYY-MM-DD.md sendo criado pelo flush
6. ✅ Sessões antigas sendo limpas após pruneDays
7. ✅ Heartbeats rodando com modelo barato (checar via /status)