What is ray-skills?

Ray Skills is a collection of three AI agent skills primarily aimed at Chinese-language content creators and product teams. The skills are designed to work with AI coding tools that support the skill or custom instruction format, including Codex, Claude Code, and Cursor. Each skill is a structured folder of workflow and reference files that an AI agent can follow step by step. The first skill, ray-content-topic-planning, takes a source document such as a whitepaper, long-form article, or interview transcript and produces a full content publishing plan from it. Given the source material and a description of the author's identity, target audience, and writing style, the skill generates around 20 candidate topic ideas, narrows them to five selected directions, produces title options and article outlines for each, and assigns scores. If historical articles are provided, it also uses them to calibrate what fits the author's established voice. The second skill, ray-multi-party-mode, simulates a structured discussion between multiple professional personas on a given question or decision. You can choose roles such as product manager, architect, QA, growth lead, or operations, and pick a discussion format: roundtable, debate, critique, planning breakdown, or decision convergence. The skill can run with one agent simulating all roles quickly, or with separate sub-agents for each role when a more independent analysis per perspective is wanted. The third skill, ray-role-creator, is a companion to the multi-party discussion skill. It creates new role definitions, such as legal, finance, sales, or recruiting personas, and outputs them in a YAML format that can be pasted directly into the role library used by the discussion skill. It checks for overlap with existing roles before creating new ones. Installation for each tool follows its own convention: copying the full skill folder into a specific local directory and restarting the AI tool. Only copying the single SKILL.md file is not enough because each skill depends on adjacent workflow and reference files.