chenkumi/codex-prompt-analysis

Codex Prompt Analysis is a small tool that looks inside a Codex coding assistant session and tells you what is eating up its prompt space before your own conversation even starts. Codex, like other AI coding tools, loads a lot of hidden context before it reads anything you type: built in tools, connected MCP servers, Codex Apps connectors, the list of available skills, and any AGENTS.md files in your project. This project measures how many tokens each of those pieces is using. It ships as a Codex skill, so you can ask Codex directly to analyze its own current prompt using a short command. You can ask it which tools or skills are taking up the most tokens, whether an MCP server or a connector is quietly adding a large schema, whether your list of available skills has grown too big, what might be safe to turn off, and whether the reported breakdown matches what your live session is actually showing. For people who prefer running things by hand, there is also a command line script you can call directly from a project that has the skill installed. It can print a plain text report, output the same data as JSON, generate a static HTML dashboard you can open in a browser, or open that dashboard automatically. You can also feed it an existing JSON report file to turn into an HTML page later. There is an optimization mode that suggests conservative ways to shrink your prompt footprint, and a relaxed mode that considers more options if the conservative pass finds nothing to trim. Overall this is a diagnostic and housekeeping tool for people running Codex sessions with many tools, skills, and MCP servers connected, meant to help you understand and control how much of the model's attention hidden context is using before your own words are read.