Pular para o conteúdo
TakoAPI
Voltar às skills

agent-audit

por sharbelayy

Coding Agents & IDEs
861Downloads informados pelo clawskills.sh
0Visualizações
0 curtidaEntre para curtir esta skill
—Ainda sem avaliações
Sua avaliação

Resumo

Audit your AI agent setup for performance, cost, and ROI.

Descrição

O que esta skill faz

Scans an OpenClaw setup to identify cost inefficiencies across agents, cron jobs, and model usage. Classifies each task by complexity, calculates per-agent and per-cron costs, and produces a report with specific downgrade recommendations and savings estimates. Works with Anthropic, OpenAI, Google, and xAI model providers.

Consolidates multi-provider cost analysis into a single command instead of manually cross-referencing each provider's billing dashboard.

Quando usá-la

  • Finding which cron jobs use expensive models for simple status checks
  • Getting a monthly AI spend estimate across all configured agents
  • Identifying tasks where model tier doesn't match actual complexity
  • Reviewing token usage patterns before scaling an agent setup
  • Generating a cost breakdown report to review AI budget allocation

Exemplo de fluxo de trabalho

ENTRADA

User asks: audit my agents and show me where I'm wasting money

AGENTE
  1. 1

    Read OpenClaw config to map all agents and their assigned models

  2. 2

    Pull cron job run history for the last 7 days, noting token counts, runtimes, and success rates

  3. 3

    Classify each task as simple, medium, or complex based on output length and reasoning requirements

  4. 4

    Calculate per-agent and per-cron monthly cost estimates using provider pricing tables

  5. 5

    Generate a markdown report listing over-provisioned tasks with specific model swap suggestions and estimated savings

SAÍDA

Markdown report with per-agent cost breakdown, flagged mismatches between task complexity and model tier, and one-liner config changes to reduce monthly spend

Instalação

clawhub install sharbelayy/agent-audit

Execute o comando da CLI ou cole o prompt no Claude Code, Cursor, Codex ou no agente de código que você usa.

Links