HRBP Skill — Usage Monitor (用量监控)

Trigger

User asks about agent usage, costs, token consumption, or resource monitoring. Examples:

• "各 Agent 用了多少？"
• "看一下本周的用量"
• "哪个 Agent 花费最多？"
• "给我看月度使用报告"

Procedure

Step 1: Clarify Query Scope

Determine what the user wants to see:

• Which agents: All agents, or specific agent(s)?
• Time range: Today, this week, this month, or cumulative?
• Metrics focus: Token usage, cost, or both?

If unclear, default to: all agents, cumulative, both tokens and cost.

Step 2: Run Usage Query

Execute the appropriate command:

# All agents, cumulative (default)
bash ./skills/hrbp-usage/scripts/agent-usage.sh

# Specific agent
bash ./skills/hrbp-usage/scripts/agent-usage.sh --agent <agent-id>

# Daily breakdown (last 7 days)
bash ./skills/hrbp-usage/scripts/agent-usage.sh --period daily

# Daily breakdown (last N days)
bash ./skills/hrbp-usage/scripts/agent-usage.sh --period daily --days 14

# Weekly breakdown
bash ./skills/hrbp-usage/scripts/agent-usage.sh --period weekly --days 28

# Monthly breakdown
bash ./skills/hrbp-usage/scripts/agent-usage.sh --period monthly --days 90

Step 3: Interpret Results

Present the data to the user with insights:

1. Overview: Total calls, total tokens, total cost across all agents
2. Per-agent breakdown: Which agents are most/least active
3. Trends: If using daily/weekly/monthly, note any patterns (increasing, decreasing, spikes)
4. Anomalies: Flag any agent with unexpectedly high usage
5. Cost efficiency: Compare input vs output tokens, cache hit ratio

Step 4: Recommendations

Based on the data, optionally suggest:

• If an agent has zero usage → ask if it should be removed
• If an agent has very high cost → suggest reviewing its model configuration
• If cache read ratio is low → the agent may benefit from prompt optimization
• If an agent hasn't been used in a long time → flag for review

Output Format

Present results in a clear, structured format:

📊 Agent 用量报告

| Agent | 调用次数 | 总 Token | 成本 |
|-------|---------|---------|------|
| main  | 150     | 500K    | $2.50|
| hrbp  | 30      | 100K    | $0.80|
| dev   | 200     | 800K    | $4.20|

总计: 380 次调用, 1.4M tokens, $7.50

趋势: 本周用量较上周增长 15%
建议: developer agent 用量最高，建议检查其模型配置

Notes

• This skill is read-only — no system modifications
• Data comes from OpenClaw session transcript files (

  ~/.openclaw/agents/<id>/sessions/*.jsonl

  )
• If no usage data exists, inform the user that agents start recording after their first interaction
• Cost data depends on model pricing configuration in openclaw.json; if pricing not configured, cost will show as "—"