CodeStable

Why we included this project

AI coding tools tend to lose their footing in long-lived codebases, where constraints and decisions from earlier work quietly drop out of context between sessions. CodeStable counters that with lightweight skill contracts for Codex and Claude Code: the model works inside explicit boundaries and has to show evidence before it claims a result, and what it learns gets written back into the project's existing docs rather than a parallel archive. It is not an agent orchestrator; a root `cs` entry routes to focused skills for features, issue fixes, refactors, epics, and read-only review, while the human keeps the final say on product decisions. Onboarding creates only a minimal project-memory skeleton, so the tooling fits into a repo without taking over its documentation, worktrees, or branch strategy. For teams who will maintain the code for years, that is what keeps AI-assisted development reviewable and reproducible.