Show HN: Curated Claude Code – a small agent harness with an intake gate (github.com)

A new tool called "Curated Claude Code" has been introduced as a compact, vetted harness for Claude Code, emphasizing quality over quantity. Unlike many repositories that prioritize sheer numbers of agents, skills, and hooks, this harness curates a small collection of effective skills designed for safe execution. It features a unique self-evolution loop, where user corrections become permanent memory and discoveries transform into verified skills, promoting a continually improving system without the risk of unverified autonomous changes. The significance of this development lies in its commitment to a disciplined approach to AI tool integration. By implementing a rigorous intake gate and a methodical evaluation process, it ensures that only relevant and secure tools are adopted, thereby preventing bloat and ambiguity common in larger frameworks. This structured design not only addresses security risks—checking for hidden instructions and potential abuses—but also establishes a feedback mechanism for learning and advancement. As self-learning models gain traction, this repository sets a new standard for creating manageable, reliable AI tools that evolve thoughtfully over time.