What is codex_audit?

Codex Audit is a set of tools and skills that help you understand why a task run through Codex, an AI coding assistant, used up so much of your five hour or weekly usage allowance. It reads the usage records Codex already keeps and breaks down where the tokens actually went, whether that is the main conversation, helper subagents, the model and effort level chosen, repeated tool output, or cached input being resent over and over. Importantly, the tool only reads and reports. It does not change your model settings, edit your configuration files, modify your codebase, or enforce any routing rules on its own. Every recommendation it makes is left for you to review and apply yourself. The repository ships three separate skills: one that audits and diagnoses usage, one that helps you delegate work to other agents with clear model and scope contracts, and one that continuously monitors a long running task and refreshes a live report. You can install and run it either by asking Codex itself to fetch and install the skill from the repository, or by cloning the repository and copying the relevant folder into your personal Codex skills directory. Once installed, you can ask Codex to audit the current task or a past task by its ID, and it will explain the biggest drivers of usage along with specific suggestions, rather than applying the same generic advice to every kind of task. The project also includes standalone Python scripts you can run directly, without needing any extra third-party packages, to analyze usage, generate a routing recommendation, or build a redacted scorecard comparing different agent configurations over time. Reports are redacted by default, hiding task titles, file paths, and usernames unless you explicitly ask for identifying detail. The README does not state a license for this project.