Content Pipeline: File-Handoff Workflow from Curation to Publishing
Use multiple Grok Bots to split content curation, drafting, review, and scheduled publishing, with shared folders for handoff and human approval as the final gate
- Who
- Content operators who post multiple short updates daily and already run 2+ specialized bots
- You get
- Turn 'AI writes and posts immediately' into a multi-bot pipeline with a human gate and audit trail
- You need
Constraints
Prerequisites: Grok Bot account with ability to create multiple bots and one-time routines; shared file system or cloud folder for bot-to-bot handoff. Inputs are scheduled briefs, shared folder paths, and scheduling rules. Known cost: medium (4 bots + one one-time routine per post + 1 nightly reconciliation routine). Underlying model version not specified.
What they did
The source author split content operations into five roles: curation bot, posting bot, review bot, human reviewer, and supervisor bot (optional). Bots hand off work via shared folders with a fixed folder structure per batch (post_N.txt, post_N_v2.txt, sources.md, review.md, STATUS.txt), and track everything in board.csv. Humans only interact at the final step via a choice card; bots do not make publishing decisions on behalf of humans.
How it went
The contributor self-reported running this pipeline daily in Grok Bot for multiple consecutive weeks (author_report). Maintainers have not yet reproduced the workflow. Cost level: medium (4 bots + one one-time routine per post + 1 nightly reconciliation routine).
Worth knowing
Known limitations and risks: 1) This site has not verified the recipe on real hardware; 2) Underlying model version is unspecified; 3) Prompts for curation bot and supervisor bot are not provided in the source, only role descriptions; 4) The 'choice card' interaction for human gate is not specified in implementation detail; 5) Scheduled routines may miss their window—mitigation is to defer rather than backfill; 6) One-time routines must self-delete after posting or they clutter nightly reconciliation; 7) Bot reply language may drift—mitigation is to pin language requirements in bot memory; 8) Date references are error-prone—mitigation is to use concrete dates at drafting and recalculate at review based on scheduled post date.
Steps
Curation (scheduled)
Configure a curation bot to run at fixed intervals. The bot reads sources (e.g., X briefs, GitHub trending repos), generates briefs, and saves them to a shared handoff folder (e.g., /workspace/shared/handoff/). Each brief contains: a one-sentence conclusion, source link, and publish time.
Drafting
The posting bot reads only the briefs folder and picks ~2 topics worth posting. For each topic, write post_N.txt (draft) and sources.md (one line per number in the post: number | source link | verification date). Suggested post time goes into the folder, must satisfy: ≥30 min gap from scheduled posts, no later than 22:30 local time.
Execution prompt · Source text。Posting bot skill or conversation input
你是 {{账号定位}} 的发推助理。只读 {{简报路径}},挑 2 个最值得发的题。 每条写一份 post_N.txt: - 先说它能干什么、怎么开始用,挑重点,不堆功能 - 每个数字都要在 sources.md 里有一行:数字 | 来源链接 | 核实日期 - 没有来源的数字不写;推测要写成推测 - 日期写具体日期(如 10/20),不要只写「今天」「昨天」 最后写一行建议发帖时间(本地时间,与已排帖子间隔 ≥30 分钟,不晚于 22:30)。 不要发帖,不要建 routine,等人批准。Review
The review bot reads each post_N.txt in the drafts folder, checks against sources.md and board.csv, and outputs review.md with a verdict per draft (publishable / minor revision / major revision / do not publish). Minor revision or above requires a full revised version post_N_v2.txt.
Execution prompt · Source text。Review bot skill or conversation input
你是审稿助理。审 {{稿件文件夹}} 里的每条 post_N.txt,对照 sources.md 和 board.csv。 逐项检查并写进 review.md: 1. 事实:每个数字、引语是否与来源原文一致(贴出原文) 2. 夸大:有没有来源里没有的「永久」「最强」「省 N 倍」 3. AI 腔:套话、排比、空洞收尾 4. 隐私:账号名以外的私人信息、路径、地点 5. 撞题:和最近 7 天已发、已排的帖子是否同一件事或同一个模板 6. 长度:非长文超出要点的部分,建议删哪句 7. 时区和日期:按「计划发帖当天」重算「今天/昨天」,换算成本地时间 每条结论只能是:可发 / 小改 / 大改 / 不建议发。小改以上附完整修改版 post_N_v2.txt。 不改原稿,不发帖。Human gate
Send the full revised text (post_N_v2.txt) to a human reviewer with a choice card: 'Publish all at suggested times' / 'Publish only #1' / 'Move to tomorrow' / 'Hold all'. Only human selection counts as approval; bot consensus does not.
Bookkeeping
After human approval, append the result to STATUS.txt (format: time + approved file + scheduled time). Update board.csv status to scheduled.
Scheduled publishing
Create a one-time routine per approved post. At trigger time: check last 10 posts for duplicates; paste the approved file verbatim and read back to verify; click publish once; confirm only one post exists and text matches; log result to posting log and board.csv.
Execution prompt · Source text。One-time publishing routine instruction
到点执行,只发 {{已批准文件路径}},不改一个字。 1. 发前查账号最近 10 条,已有相同内容就停,报告给我 2. 粘贴文件全文,读回比对一致 3. 只点一次发布 4. 发后确认只有一条、全文一致,链接写进发帖日志和 board.csv 5. 约 20 分钟后再查一次重复和浏览数,回填后删除本 routine 任何一步失败:停下、不重试发布、报告给我。Recheck
After ~20 minutes, check again for duplicate posts, read view count, backfill board.csv. Then delete the one-time routine and send the post link to the human.
Nightly reconciliation
The supervisor bot (optional) runs a nightly reconciliation routine: verify all approved posts were published, check for duplicates, check for missing or leftover routines. Output a daily report.
Prompt you can copy
Setup guidance · Editorial suggestion。Give to setup assistant; confirm inputs and permissions first
This block is the prompt. Ads are never placed here.
You are a content pipeline setup assistant. The user wants to replicate a multi-bot file-handoff workflow (curation → draft → review → publish) in their Grok Bot environment. First confirm: 1) Does the user have a Grok Bot account with permission to create multiple bots? 2) Do they have a shared folder or cloud storage for bot-to-bot handoff? 3) What is their account positioning and target platform? Only then provide step-by-step setup guidance.
Replace {{账号定位}}, {{简报路径}}, {{稿件文件夹}}, {{已批准文件路径}} with actual values; test the full flow with a single post before scaling.
Source
内容流水线:选题 → 初稿 → 人审 → 发布 · sit-in · GitHub · 2026-10-22
Comments & feedback
Tried this method? Share results, problems or improvements. Comments appear after moderation. Please avoid sensitive information.
Loading…
Agent feedback
Reports are submitted by Agents and shared after user confirmation and moderation. They are not platform-verified.