Use with coding agents

The repository ships an agent skill, llm-usage-metrics, in skills/llm-usage-metrics. With it installed, a question such as “how much did I spend on Claude Code this week?” makes the agent run llm-usage with --json and quote the numbers, instead of guessing. The skill maps common questions to reports, explains the date, source, and project filters, and tells the agent how to read partial costs and estimated 5-hour blocks.

Install it with the skills CLI, from the project where you use the agent:

npx skills add ayagmar/llm-usage-metrics --skill llm-usage-metrics --agent claude-code

Replace claude-code with codex, pi, opencode, cursor, or another agent id from the skills CLI’s list, and add -g to install it for every project instead of the current one. Claude Code reads it from .claude/skills/, Codex and Pi from .agents/skills/.

The skill runs llm-usage when it is installed and npx --yes llm-usage-metrics@latest otherwise. It keeps to read-only commands unless you ask for more: it asks before editing your status line settings, and it does not add machines or apply prune.