菜单
Summarize the current conversation: ongoing task progress + continuous improvement insights. Use when the user says "summarize this", "recap", "summary", "wrap up", or wants a checkpoint of work-in-progress and lessons learned.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Extract content from an X/Twitter post, thread, or article using Playwright. Detects threads automatically, walks back to the first post when the shared link lands mid-thread, and downloads images locally. Use when the user says "read this x post", "get content from this tweet", "what does this x post say", "extract this tweet", "get the whole thread", or shares an x.com/twitter.com URL.
This skill should be used when the user shares a blog/article URL (Medium, Substack, personal blog, news/long-form post) and asks to "download this blog post", "save this article", "archive this post with images", "clip this medium article", or otherwise wants a self-contained local copy. Extracts clean markdown via Defuddle, recovers lazy-loaded images via Playwright (which Defuddle drops), and — when the article contains Vedic astrology charts — invokes the download:vedic-chart skill to digitize them. Saves a folder with frontmatter + an images/ subfolder.
This skill should be used when the user shares a Vedic/Jyotish horoscope chart image and asks to "convert this chart", "extract this kundli", "digitize this horoscope", "chart to ascii", "read this birth chart", "what planets are in this chart", or to parse a rasi/navamsa/divisional (D-1, D-9, D-7, D-30, etc.) chart printout. Converts a chart image into structured JSON plus ASCII North-Indian and South-Indian diagrams. Also invoked by the download:blog-post skill when astrology charts are found in an article.
Parallel divergent ideation for coding agents. Spawns N isolated branches under different cognitive frames (regulator, biology, speedrunner, 10 year old, $0 budget), scores, prunes traps, and deepens the top survivors. The isolated parallel branches and the separated generator/critic phases are load-bearing. Costs about 10 LLM/Agent calls per run (5 to 10x a single answer) so invocation is gated. UNCONDITIONAL TRIGGER. When the user types "/adhd" or explicitly asks for "ADHD mode", "use the adhd skill", or "run ADHD on this", invoke the full loop without further judgment. Skip the pre-flight check. CONDITIONAL TRIGGER. For everything else, self-judge before invoking. Only proceed when ALL of these hold: (a) the answer space is OPEN-ENDED (multiple viable answers, no single canonical correct one), (b) the cost of the obvious answer being wrong is high (architecture decision, fuzzy bug with no known root cause, API or SDK surface design, naming for a public product, strategy, positioning, schema
Propose the single smartest, most radically innovative, accretive, useful, and compelling addition to the current project. Output uses BLUF + Pyramid Principle. Use when the user says "next idea", "what should I build next", "what's the best next move", "what would make this 10x better", or wants a single high-conviction recommendation grounded in the current state of the project.
Check whether a candidate company name is available across USPTO trademarks, WA Secretary of State, and major domain TLDs. Use when the user asks to check, verify, clear, or sweep a name for an LLC, corp, brand, trademark, or domain — anything that needs to be unique before filing.
| name | summarize-this-session |
| description | Summarize the current conversation: ongoing task progress + continuous improvement insights. Use when the user says "summarize this", "recap", "summary", "wrap up", or wants a checkpoint of work-in-progress and lessons learned. |
Review everything that has happened in this conversation and produce a two-section summary.
Look back through the entire conversation history — every user message, every tool call, every result, every correction.
Produce the following two sections:
Summarize the work in progress:
Analyze the conversation for learning opportunities:
Mistakes Made by Claude: Actions Claude took that were wrong, suboptimal, or had to be redone. Include:
Mistakes Pointed Out by Human: Corrections the user explicitly made. Include:
Lessons Learned: Actionable takeaways from this conversation that would improve future sessions. These could be:
