oil-oil/codex-explore-skill

This is a small add-on for Codex, an AI coding assistant. The add-on is called a skill, which is a piece of behaviour you can plug into Codex so the assistant follows specific rules when a particular keyword is mentioned in the conversation. This particular skill is named Explore. The job of Explore is to make Codex stop and look around before it starts changing code. When a developer is about to ask Codex to work on a project, especially one Codex has not seen before, it can be slow and noisy if the main assistant reads dozens of files itself. Instead, the Explore skill tells Codex to first hand the reconnaissance work off to smaller helper agents called explorer subagents. Those helpers read the project, take notes, and report back a short summary plus a table of the most important files and suggestions for what to read next. The main conversation can then concentrate on planning the actual change instead of being filled up with raw source code. Installing the skill is one shell command: clone the repository into the ~/.codex/skills/explore folder. After that, the user mentions $explore in a Codex prompt, for example "Use $explore to map this codebase before making changes," and the skill takes over. The README lists the rules the skill enforces. The main agent is not allowed to read the target codebase while the explorer subagents are still working. Before any large local reading or editing begins, a key-files table must exist. The explorers are asked to label files as primary, legacy, experimental, generated, unused, or unclear when they can tell, and to group likely changes by concern such as user interface, API, state management, data model, permissions, internationalisation, tests, or configuration. The project is released under the MIT licence. The README is short and does not list a programming language or any additional configuration options.