what is feature-track fr?

Feature Track is a lightweight, repo-native way to keep AI coding agents like Codex and Claude Code aligned on what each long-lived feature in a project actually is, right now. Instead of a chat history search or a database, it is a set of Markdown files that live inside your own repository under docs/features, one per feature plus a global index. The problem it addresses is that AI coding agents are good at making changes but bad at holding onto context. Feature knowledge tends to be scattered across product docs, API notes, plans, old chats, and commits, so agents often rediscover the same things repeatedly, and stale plans can look authoritative even after the real system has moved on. Switching between tools or sessions usually resets whatever memory an agent had built up. Feature Track gives each important feature a small living file that records its current status, which docs are still the real source of truth, key decisions, risks, and recent changes, without replacing the existing documentation it points to. It is aimed at teams and solo developers who want shared project memory without a heavy process, especially brownfield projects with useful but scattered documentation, and anyone using Codex, Claude Code, or several agent tools together. The design favors linking to existing docs first and migrating content later, so it can be adopted gradually rather than requiring a full rewrite of a project's documentation. Installation is a single command, npx skills add JunsW/feature-track, with alternate paths for installing it as a Codex plugin, a Claude Code plugin, or through local Python adapters. A companion CLI can initialize the feature index, add new feature tracks, and validate that the index has not drifted, required sections are present, and internal links are not broken. Validation is intentionally lenient by default so it can be tightened later in continuous integration. Feature Track describes itself as lighter than full spec-driven development frameworks, since it focuses on remembering what a feature is rather than governing every step of building it. The project is licensed under the MIT license and is currently at version 0.1.0, its first public release.