wtf is codex-hygiene?

Codex Hygiene is a skill for OpenAI's Codex that helps you understand how you have been using it. It is read only and does not change any settings on its own, it just measures and reports. There are two main ways to use it. The first is a quick measurement that prints small counts pulled from recent Codex telemetry, such as tool list activity and token usage per conversation thread. The second is a fuller activity review covering a chosen window of one to ninety days, which can be output as Markdown or as machine readable JSON. This review pulls from local SQLite telemetry and, by default, digs a bit deeper into rollout records to add details like how long tasks took, how much output different tools produced relative to each other, how often verification commands were run, when context was explicitly compacted, and when skill documentation was read. The review also looks at your current Codex profile: which plugins are installed and their state, whether MCP connections and Apps are available, and how many project specific configuration stanzas exist. After gathering all of this, it separates what it directly observed from what it is interpreting, and it clearly marks anything it cannot determine as unknown. Based on the measurement, it can suggest small, reversible cleanup steps, but it will not make any changes without your explicit approval. Privacy is treated carefully. Full logs, configuration files, tool schemas, secrets, and environment variable values are never included in a report, and prompts, responses, thread titles, and file paths stay private too. To install it, you clone the repository into your local Codex skills folder, and Codex should detect it automatically. It works on macOS and Unix like systems and needs common shell tools plus Python 3 for the fuller review. It is released under the MIT license.