Agent Audit — 代理 审计

代理 审计

扫描 your entire OpenClaw 设置up and 获取 actionable cost/performance recommendations.

What This 技能 Does 扫描s config — reads OpenClaw config to map 模型s to 代理s/tasks Analyzes cron 历史 — 检查s every cron job's 模型, 令牌 usage, 运行time, 成功 rate Classifies tasks — determines complexity level of each task Calculates costs — per 代理, per cron, per task type using 提供者 pricing Recommends changes — with confidence levels and risk 警告s 生成s 报告 — markdown 报告 with specific savings estimates 运行ning the 审计 python3 {baseDir}/scripts/审计.py

Options:

python3 {baseDir}/scripts/审计.py --格式化 markdown # Full 报告 (default) python3 {baseDir}/scripts/审计.py --格式化 summary # Quick summary only python3 {baseDir}/scripts/审计.py --dry-运行 # Show what would be analyzed python3 {baseDir}/scripts/审计.py --输出 /path/to/报告.md # Save to file

How It Works Phase 1: Discovery Read OpenClaw config (~/.OpenClaw/OpenClaw.json or similar) 列出 all cron jobs and their configurations 列出 all 代理s and their default 模型s 检测 提供者 (Anthropic, OpenAI, Google, xAI) from 模型 names Phase 2: 历史 Analysis Pull cron job 运行 历史 (last 7 days by default) Calculate per-job: avg 令牌s, avg 运行time, 成功 rate, 模型 used Pull 会话 历史 where avAIlable Calculate total 令牌 spend by 模型 tier Phase 3: Task Classification

Classify each task into complexity tiers:

Tier Examples Recommended 模型s Simple 健康 检查s, 状态 报告s, reminders, 通知 Cheapest tier (HAIku, GPT-4o-mini, Flash, Grok-mini) Medium Content drafts, re搜索, summarization, data analysis Mid tier (Sonnet, GPT-4o, Pro, Grok) Complex Coding, architecture, security review, nuanced writing Top tier (Opus, GPT-4.5, Ultra, Grok-2)

Classification 签名als:

Simple: Short 输出 (<500 令牌s), low thinking requirement, repetitive pattern, 状态/健康 tasks Medium: Medium 输出, some reasoning needed, creative but templated, re搜索 tasks Complex: Long 输出, multi-step reasoning, code generation, security-critical, tasks that previously fAIled on weaker 模型s Phase 4: Recommendations

For each task where the 模型 tier doesn't match complexity:

⚠️ RECOMMENDATION: 降级 "Knox 机器人 健康 检查" from opus to hAIku Current: anthropic/claude-opus-4 ($15/M 输入, $75/M 输出) Suggested: anthropic/claude-hAIku ($0.25/M 输入, $1.25/M 输出) Reason: Simple 状态 检查 averaging 300 输出 令牌s Estimated savings: $X.XX/month Risk: LOW — task is simple pattern matching Confidence: HIGH

Safety Rules — NEVER Recommend Downgrading: Coding/development tasks Security reviews or 审计s Tasks that have previously fAIled on weaker 模型s Tasks where the user explicitly chose a higher 模型 Complex multi-step reasoning tasks Anything the user flagged as critical Phase 5: 报告 Generation

输出 a 清理 markdown 报告 with:

Overview — total 代理s, crons, monthly spend estimate Per-代理 breakdown — 模型, usage, cost Per-cron breakdown — 模型, frequency, avg 令牌s, cost Recommendations — 排序ed by savings potential Total potential savings — monthly estimate One-liner config changes — exact 模型 strings to swap 模型 Pricing Reference

See references/模型-pricing.md for current pricing across all 提供者s. 更新 this file when prices change.

Task Classification DetAIls

See references/task-classification.md for detAIled heuristics on how tasks are classified into complexity tiers.

导入ant Notes This 技能 is read-only — it never changes your config automatically All recommendations include risk levels and confidence scores When unsure about a task's complexity, it defaults to keeping the current 模型 The 审计 should be re-运行 periodically (monthly) as usage patterns change 令牌 counts are estimates based on cron 历史 — actual costs depend on your 提供者's billing