Research idea evaluation and iteration assistant covering the six-stage LLM Wiki pipeline — paper discovery (arXiv/Semantic Scholar), ingest to raw/, Claude Code compiles raw/ to a structured wiki/, Obsidian visualization, multi-model debate (Claude/Gemini/Codex CLI), and Perplexity novelty validation. Use when the user wants to register a new idea, compile a paper wiki, run a multi-model debate, or validate an idea's novelty. Triggered by keywords like "idea validate", "paper wiki", "compile raw", "novelty check", "多模型讨论", "idea 验证", or direct invocation.
Build a comprehensive Chinese paper card under research-workspace/ideas/<slug>/wiki/papers/<paper>.md — one card that lets reader fully understand the paper in a single read without reopening the PDF. Covers TL;DR + 背景与任务 + 方法概览 + 方法细节（含公式直觉） + 实现 + 关键结果（表 + 解读）+ 关联工作 + 与本课题的关系. Faithful prose explanation with original symbols preserved. **No critical analysis, no judgment, no gap-finding, no author affiliations, no future plans**. Critical reading happens in AI Assistant chats, not in the card. Embeds PDF reader iframe at top. Triggered by "论文卡片", "paper card", "写一张卡片", "新增论文卡片", or direct invocation.
对学术论文做批判精读 + 研究缺口分析。区分作者声称 vs 真实成立 vs 未明说，并从用户研究背景出发找可切入的研究空间。当用户上传或粘贴一篇论文（PDF / arXiv 链接 / 纯文本），或要求"精读 / 讲解 / 总结 / limitation 在哪 / 这篇怎么样"时使用。同方法论同时被右侧 AI 助手用于论文卡片页对话——canonical 方法论存在 `research-workspace/methods/paper-critical-reading.md`，本 skill 是 Claude Code 入口。