Show HN: Token-saviour – agent skill from benchmarking 9 token-saving tools (github.com)

A new AI agent skill called "token-saviour" has been introduced, enhancing efficiency in coding tasks by intelligently routing operations to the most token-efficient tools. This skill benchmarks nine different token-saving tools across four distinct layers of token consumption—code reading, command output, prose generation, and code generation. The results indicate significant savings, with the best combinations achieving up to 69.6% reduction in token usage when employing the optimal tools like serena, rtk, and caveman for specific tasks. This development is significant for the AI/ML community as it directly addresses the high token costs associated with coding operations, providing a streamlined approach that minimizes unnecessary overhead. With explicit rules to classify tasks and reject anti-patterns, token-saviour enables developers to manage resource allocation effectively. The tool's ability to gracefully degrade to basic operations when certain tools aren't available ensures broad usability, making it a practical addition to developers’ workflows. The findings are backed by thorough benchmarks on realistic Python codebases, reinforcing the utility and credibility of the token-saviour agent skill.