| name | aispark-daily-grok |
| description | Daily curated highlights from the AISpark community on X. Fetches recent posts from the 11 key members (aehyok, MinLiBuilds, Lonely__MH, cheery9998, etc.), excludes replies/comments, **hard-filters to strictly the last 24 hours by tweet ID** (X's since: is calendar-day granular and over-captures, so we drop anything older than 24h), internally classifies types (单纯的post / 文章 / 引用), selects up to 8 valuable shares (tools, workflows, resources, insights, tutorials), and outputs up to 8 lines in the strict format: 序号、本社群 昵称分享:精炼标题 链接. Slash command: /aispark-daily-grok. Triggers on: aispark-daily-grok, 社群日报, AISpark daily, 本社群24小时, 本社群分享, 社区精选, 24h posts, 跑一下社群, community highlights, AISpark 精选.
|
aispark-daily-grok
When this skill is invoked (by /aispark-daily-grok or matching intent), immediately execute the workflow below and return ONLY the final list (up to 8 lines). Do not add any extra text, explanations, or markdown unless the user explicitly asks for more details.
User Mapping (exact nicknames)
Use these exact nicknames in the output:
- MinLiBuilds → 实践哥
- Lonely__MH → L总
- cheery9998 → cheery老师
- xiangxiang103 → 雨哥老师
- legacyvps → 小墨老师
- alin_zone → 阿蔺老师
- LawrenceW_Zen → 劳伦斯老师
- Liu_zhongxisn → Drunk老师
- ModengSir → 优秀学员勇哥
- MainCaseZ → 优秀学员张总
- jinglian → 前端哥老师
- aehyok → AI少年老师
Step-by-step Workflow
-
Determine the time window (and the 24h cutoff tweet ID)
- X 的
since: operator 是按自然日的,不是滚动 24 小时——傍晚/夜里跑时,since:昨天 实际会捞进约 47 小时的帖子。所以我们用 since: 多捞当网,再按推文 ID 硬过滤到严格的近 24 小时。
- 取一个「24 小时前」对应的最小推文 ID
MIN_ID,凡是 status ID < MIN_ID 的帖子都超过 24 小时,一律丢弃。
- 首选:宿主(Claude 桥接器)会在 prompt 里给出「现在时间」和一个具体的
MIN_ID。给了就直接用。
- 兜底(有人不经桥接、直接在 grok 里跑时自己算):用 Twitter snowflake 公式。
- snowflake 起点
EPOCH = 1288834974657(ms)。
- 24 小时前的毫秒数
m = 现在的 Unix 毫秒 - 86400000。
MIN_ID = (m - EPOCH) << 22(即 (m - 1288834974657) * 4194304)。
- 反过来,任一帖子的发帖时间
ms = (ID >> 22) + EPOCH,可用来核对。
since: 仍用昨天的日期(since:YYYY-MM-DD)即可,宁可多捞,过滤交给 MIN_ID。
-
Fetch posts for all members
-
Process each post and extract data
- From the tool result, extract:
- Post ID (the number after "ID:") — 这个 ID 既用来建链接,也用来做 24h 硬过滤,务必保留。
- Author handle (use the one from the query or the @username in "Author:")
- Content (the full text after "Content:")
- Engagement numbers (Likes, Views, etc.)
- Whether a "Quoted Post:" section exists
- Any Media or links mentioned
- Construct the full URL:
https://x.com/{handle}/status/{ID} (use lowercase handle).
- Classify type internally (do not output the type in the final list):
- If the result contains a "Quoted Post:" block → "引用"
- Else if Content is long (> ~350 chars), contains numbered lists, GitHub links, detailed steps, "教程"/"指南"/"资源"/"方法论", or X article links (
/i/article/) → "文章"
- Otherwise → "单纯的post"
-
Hard-filter to strictly the last 24 hours
- 逐条比较:只保留
status ID >= MIN_ID 的帖子;ID < MIN_ID 的(超过 24 小时)直接剔除,不进入候选池。
- 这一步是这个 skill 的硬约束:绝不为了凑满 8 条而把超过 24 小时的帖子放回来。宁可少,也不要把过期的塞进来。
- 过滤后若候选寥寥(比如只剩三五条),那就是这 24 小时社群确实发得少——如实输出实际条数。
-
Curate up to 8 items
- 只在第 4 步过滤后的帖子里挑。
- Filter out very low-value posts (e.g. < 3 likes + < 200 views and purely casual one-liners).
- Rank and select the best (最多 8 条) using these priorities (in order):
- High engagement (Likes + Views + Bookmarks).
- Practical value to the community: real tool usage, workflows, coding tips, resource lists, tutorials, deep insights, "保姆级" guides, indie hacking / AI building experiences.
- Diversity: spread across different members when possible. Avoid picking multiple quotes of the exact same post unless the commentary adds unique value.
- Presence of media, external links (GitHub, articles, tools), or structured content.
- 数量规则:目标最多 8 条;若 24h 内达标的帖子不足 8 条,就输出实际条数(哪怕只有 4 条),不要用超过 24 小时的帖子补足。
-
Create the title / summary for each selected post
- Turn the post Content into a short, attractive, refined Chinese title or one-sentence summary.
- It should read naturally after "分享:".
- Keep it concise (ideally under 25-30 characters when possible), capture the core value or hook.
- Examples of good style:
- "零基础 Vibe Coding 推荐的 5 个优质开源项目"
- "用 Helio 搭建自媒体 AI 工作流团队的实战经验"
- "DeepSeek 对接 Codex 的正确打开方式 + 视频演示"
-
Map handle to nickname and format output
Important Rules
- 严格近 24 小时:每条入选帖子必须
ID >= MIN_ID。这是最重要的约束——since: 捞回来的超 24h 帖子必须被第 4 步剔除。
- Never include replies or comments.
- Never output the internal post type in the final list (user only wants the clean formatted list).
- Be consistent with the exact output format the user specified.
- Use only the provided X search tool for recency and accuracy.
- If a post is a quote, you may still select it if the author's commentary adds strong value.
- Prioritize quality over quantity — the selected items should feel like the "best of" the community's last-24h activity;条数少是可以接受的,过期混入不可以。
Execute the full workflow as soon as the skill is triggered and return only the list (up to 8 lines).