一套可重复使用的AI代理技能合集，供个人使用。该仓库维护Codex、Claude Code及其他兼容技能格式代理的说明、工作流程及支持资源。

仓库结构

每个技能都存在于其顶层目录中。目录名称必须与其 SKILL.md 中的 name 相匹配：

除了SKILL.md，只有在技能真正需要时才添加agents/、references/、scripts/或assets/。

跨平台兼容性

• SKILL.md 是技能的跨平台核心。它定义了触发条件和主要工作流程。
• [references/、scripts/ 和 assets/ 包含仅在需要时加载的辅助资源。
• agents/openai.yaml 是 Codex/OpenAI 产品中可选的元数据，用于显示名称、简短描述、默认提示及其他界面功能。
• Claude Code 不依赖于 agents/openai.yaml;该文件既不替代也不更改 SKILL.md 中的通用指令。
• 技能可以包含多个代理平台的可选适配器，但其核心工作流程不应依赖于任何单一平台。

技能

加速领域学习

Skill

A collection of reusable AI Agent Skills for personal use. This repository maintains instructions, workflows, and supporting resources for Codex, Claude Code, and other agents compatible with the Skills format.

Repository Structure

Each Skill lives in its own top-level directory. The directory name must match the name in its SKILL.md:

skill/
├── README.md
└── /
    ├── SKILL.md
    ├── agents/
    │   └── openai.yaml
    ├── references/
    ├── scripts/
    └── assets/

Other than SKILL.md, add agents/, references/, scripts/, or assets/ only when the Skill actually needs them.

Cross-Platform Compatibility

• SKILL.md is the cross-platform core of a Skill. It defines triggering conditions and the primary workflow.
• references/, scripts/, and assets/ contain supporting resources loaded only when needed.
• agents/openai.yaml is optional Codex/OpenAI product metadata for display names, short descriptions, default prompts, and other interface features.
• Claude Code does not depend on agents/openai.yaml; this file neither replaces nor changes the universal instructions in SKILL.md.
• A Skill may include optional adapters for multiple agent platforms, but its core workflow should not depend on any single platform.

Skills

accelerated-domain-learning

Accelerated Domain Learning helps learners build a reliable mental map of an unfamiliar field quickly, then tests genuine understanding through evidence checks, deep questions, and adaptive follow-ups.

Key capabilities:

• Calibrate the learning goal, existing knowledge, and observable completion criteria.
• Establish domain boundaries, prerequisites, mental models, and concept relationships.
• Distinguish consensus, major debates, open questions, and frontier directions.
• Select evidence sources by knowledge type and label facts, opinions, and inferences.
• Generate explanation, distinction, boundary, counterexample, transfer, and evidence-evaluation questions.
• Locate misconceptions, repair missing prerequisites, and adapt the depth based on the learner’s answers.
• Produce a cognitive summary and continued learning path in the learner’s own language.
• Support scientific, technical, humanities, professional-practice, and general-knowledge domains.

Directory: accelerated-domain-learning

prd-writer

Universal PRD Writer turns product ideas, business needs, feature requests, and existing specifications into product requirements that can be reviewed, accepted, and implemented.

Key capabilities:

• Use one requirements core for standard digital products, AI products, Agent products, and multi-Agent systems.
• Adjust output depth based on audience, product type, requirement size, and deliverable.
• Define the problem, users, goals, scope, flows, features, metrics, risks, and acceptance criteria consistently.