菜单
Extract and summarise content from URLs, YouTube videos, podcasts, and files. Use when the user shares a link, asks "what's this about", "summarise this", wants a transcript, or references web/video/podcast content.
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
Publish dynamic web pages for dashboards, trackers, or any data viewable in a browser. Server-side Python runs on every request.
Fetch X/Twitter timeline and bookmarks. Use when the user asks about their Twitter feed, "what's on my timeline", "my bookmarks", "what have I bookmarked", or wants raw tweet data.
General-purpose lists — shopping, questions for appointments, packing lists, etc. Use when the user wants to create a list, add items, "shopping list", "questions for appointments", "what's on my list", or manages grouped items around a theme or event.
Manage reminders — one-off and recurring. Use when the user says "remind me", "set a reminder", "what reminders do I have", "cancel that reminder", or needs time-based nudges. Wraps the cron system with history and categories.
Personal knowledge base using the Zettelkasten method. Use when the user wants to save a note, look up knowledge, connect ideas, "what do I know about X", "save this", "note this down", or asks about their notes/knowledge base.
Query Garmin Connect health and fitness data including activities, sleep, heart rate, HRV, stress, and body battery. Use when the user asks about their Garmin data, "how did I sleep", "resting heart rate", "HRV", "stress levels", "body battery", "training status", "Garmin activities", or wants to analyze health metrics from Garmin Connect.
| name | summarise |
| description | Extract and summarise content from URLs, YouTube videos, podcasts, and files. Use when the user shares a link, asks "what's this about", "summarise this", wants a transcript, or references web/video/podcast content. |
Extract clean content or generate summaries from URLs, YouTube videos, podcasts, and files using @steipete/summarize via buns.
buns run --allow-host="*" --allow-env="*" --memory=512 --timeout=120 workspace/skills/summarise/scripts/summarize.ts -- [FLAGS] "URL_OR_FILE"
Use --extract to get clean text, then summarise as the agent. This avoids a redundant LLM call.
# Web page or article
... -- --extract "URL"
# YouTube with timestamps
... -- --extract --timestamps "YOUTUBE_URL"
When content is very long and would use too many tokens, let summarize do it directly.
... -- "URL" # default xl summary
... -- --length short "URL" # brief summary
--extract for most use cases — the agent can summarise in its own voice--extract) for very long content to save tokens--timestamps adds time markers to YouTube/podcast transcripts--json returns structured output with metadata--timeout for long videos or slow sitesANTHROPIC_API_KEY)