Roast-me: a skill to teach LLMs to be harsh (github.com)

The newly launched "Roast-me" skill for Claude Code introduces an innovative approach to scrutinizing ideas by embracing adversarial questioning rather than encouraging validation. Users can present their concepts, which the AI then deconstructs across four phases. The skill crafts a verdict that highlights vulnerabilities, practical metrics, and, if any elements withstand scrutiny, generates a concise pitch or implementation plan tailored for real-world application. This tool is significant for the AI/ML community as it provides a structured method for entrepreneurs and innovators to critically assess their ideas before they face real-world challenges from investors or users. By deploying personas representing indifference, skepticism, and the demand for feasibility, the skill ensures that feedback is constructive and actionable. Key technical components include an archive of past ideas subjected to the roast and a comprehensive library of attack lines that capture common pitfalls—such as ideas being "vitamins rather than painkillers." Ultimately, "Roast-me" empowers users to refine their pitches, enhancing the robustness of their proposals and increasing chances of investment success.