A good build-in-public thread isn't written, it'sdug out of real experience. This skill does the grunt work of digging and assembling; the tone-setting words and the sharp edge of opinion are up to the author.
Iron rule: no ghostwriting
The output is alwaysskeleton + materials, not a finished piece:
- Facts, numbers, timeline, code/commands, screenshot slots—fill them in solid
- Opinion sentences, judgment sentences, emotion sentences—leave
【你的话:____】blank slots, each with a one-line hint of "what the reader is waiting for here" - Forbidden to generate any opinion statement in the author's first-person voice
Violate this, and the entire output is void.
Input
All three are accepted:
/ray-thread 上周排查并清掉了几台服务器上的挖矿木马,想发条 thread
/ray-thread ~/projects/xxx/RETRO.md # 复盘文档
/ray-thread (无参数) # 追问最近有什么值得讲的实战
When information is insufficient, ask follow-ups, all at once: What was the starting point of this? The worst moment? The final numerical result? Any counterintuitive findings?
Step 1: Material archaeology
Dig five types of ore from the raw material, list them as a material inventory:
| Ore | What to look for | Why it's valuable |
|---|---|---|
| Hard numbers | Amounts, time spent, quantities, percentages, before-and-after comparisons | The credibility anchor of the thread; a thread without numbers is prose |
| Turning point | The "thought it would be... turned out..." moment | The engine of the narrative |
| Failures and costs | Wrong paths taken, money wasted, things lost | The trust currency of build-in-public—only showing wins is advertising |
| Counterintuitive conclusions | Findings that run counter to common sense/tutorials/what most people do | The reason to repost |
| Reusable actions | Concrete steps/commands/checklists readers can copy today | The reason to bookmark |
Dig at least 1 item per type; for types you can't dig out, mark "missing" and prompt the author to fill it in (the gap itself may mean this experience isn't ready for a thread yet).
Step 2: Structure selection
Choose a skeleton based on the shape of the material, pick only one:
- Incident postmortem type(has failures and costs): calm opening → what happened → worst moment → handling → numerical result → lessons checklist
- Decision rewind type(has a turning point): conclusion first → rewind to the decision scene → the options at the time → why this one was chosen → looking back now
- Open ledger type(has hard numbers): all numbers revealed → break down the composition → the most surprising item → next-stage goal
- Method distillation type(has reusable actions): pain-point scenario → my approach broken down step by step → the pitfall in each step → wrap up with a complete checklist
- Arc narrative type(long time span): starting state → three key turns → current state → one sentence summing up this stretch of the road
Write one line for the rationale: "Chose type X, because the strongest thing in the material is ____".
Step 3: Tweet-by-tweet assembly
Output a tweet-by-tweet skeleton, each tweet containing:
T3 ─────────────────────
[事实] 那天发现 /tmp 下有挖矿进程,htop 一看 CPU 拉满,多台机器全中
[你的话:____] ← 这里读者在等:你发现那一刻的第一反应
[配图位] htop 截图(如有)
Assembly discipline:
- Each post ≤ 240 characters (English) / ≤ 100 characters (Chinese); split if over
- T1 is the hook (see Step 4); T2 must deliver on T1's promise, no delaying to T3
- Put numbers at the front of the post whenever possible
- 7-12 posts across the thread; if the material can't support 7 posts, suggest switching to a single post instead, don't force it
- Last post: summary + an open question or a preview of next steps (leave a hook for replies)
Step 4: Hook candidates
Give 3-5 candidates for T1, across different levers:
- Number lever: put the most brutal number in the first sentence
- Loss lever: lead with the cost ("this mistake cost me X")
- Contrast lever: common belief vs my finding
- Suspense lever: result first, process after
Label each candidate with the psychology it's betting on.Don't choose for the author; leave the ranking to the author.
Step 5: Publishing notes
- Suggested posting time (by target audience time zone)
- List of images to prepare (screenshots/charts, note which image goes with which post)
- One-line reminder: which post is most likely to be taken out of context; self-check before publishing
Quality self-check
- Are all five categories of the material checklist present? Are any missing ones explicitly flagged?
- Did any viewpoint written on the author's behalf sneak into the skeleton? If yes → delete it and replace with a blank
- Read all the [facts] together: can a stranger understand the whole story of this matter? If not → the skeleton has holes
- If this thread goes out, what's the best outcome? (gain followers/drive traffic/build a professional image/spark discussion) — confirm with the author before finalizing