O que audit-agent faz?
Audits AI-agent runs and session snapshots for token usage, cache traffic, cost, duration, turns, tool calls, failures, retries, subagent attribution, duplicated reads, and artifact yield. Use when reporting or comparing a Codex CLI rollout, GitHub Copilot CLI, Gemini CLI, OpenCode, Claude Code, a runner export, or transcript logs. Do NOT use to monitor live CI, prove a check can fail, or sandbox a CLI.