name: skill-cost version: 0.1.0 author: dzwalker description: > Track per-skill token usage and costs from OpenClaw session logs. Use when user asks about skill-level spending, which skill costs the most, or wants a per-skill cost breakdown. IMPORTANT: This is a bash-tool skill. You MUST use the bash/shell tool to execute commands. triggers:

• skill cost
• skill spending
• skill usage
• which skill costs
• per-skill cost
• skill token
• cost by skill
• token by skill
• skill breakdown
• cost breakdown by skill repository: https://github.com/dzwalker/skill-cost license: MIT dependencies: [] tools:
• skill_cost_report
• skill_cost_detail
• skill_cost_compare metadata: {"clawdbot":{"emoji":"💰","requires":{"anyBins":["python3","bash"]},"os":["linux","darwin","win32"]}}

Skill Cost — USE BASH TOOL

You MUST use the bash or shell tool to run these commands. No other method works.

Commands

Per-skill cost report:

bash ~/.openclaw/workspace/skills/skill-cost/skill-cost.sh report

Last 7 days:

bash ~/.openclaw/workspace/skills/skill-cost/skill-cost.sh report --days 7

Since a specific date:

bash ~/.openclaw/workspace/skills/skill-cost/skill-cost.sh report --since 2026-03-01

JSON output:

bash ~/.openclaw/workspace/skills/skill-cost/skill-cost.sh report --format json

Top skills by cost:

bash ~/.openclaw/workspace/skills/skill-cost/skill-cost.sh ranking

Detail for a specific skill:

bash ~/.openclaw/workspace/skills/skill-cost/skill-cost.sh detail poe-connector

Compare two skills:

bash ~/.openclaw/workspace/skills/skill-cost/skill-cost.sh compare poe-connector web-search

DO NOT

• Do NOT use sessions_spawn — it will fail
• Do NOT use browser — it will fail
• Do NOT use any approach other than bash tool — only bash works

How It Works

1. Scans OpenClaw session JSONL files at ~/.openclaw/agents/*/sessions/
2. Parses assistant messages for tool calls and token usage
3. Attributes token usage to skills by matching bash command paths and tool names
4. Aggregates and reports per-skill costs with model and daily breakdowns

Notes

• No API keys required — reads directly from local session files
• No external Python dependencies (stdlib only)
• Skill attribution uses bash command paths and SKILL.md tool mappings
• Tokens for multi-skill messages are proportionally split across skills
• Usage not attributable to any skill is categorized as (built-in) or (conversation)