| name | openmark-newsletter-thread |
| description | Compose a short LinkedIn-post-style thread from Ahmad's OpenMark saves. Use when Ahmad wants a tight 200–400 word post he can paste into LinkedIn, X, or a short newsletter slot — NOT a long essay or full digest. Trigger phrases include "thread", "LinkedIn post", "short post", "tweet thread", "quick newsletter", "social post", "/newsletter-thread". Output is one strong hook + 4–7 short paragraphs + ONE link. Optimized for skim-reading on a phone. |
| metadata | {"type":"composition"} |
OpenMark — Newsletter Thread
Short. Punchy. Phone-readable. ~250-400 words total. ONE link. No tables. No bullet lists.
When to use vs other newsletter skills
| User intent | Skill |
|---|
| "Quick LinkedIn post" / "short post" / "tweet" | this one (thread) |
| Full newsletter, multiple sources | openmark-newsletter |
| Categorical recap | openmark-newsletter-roundup |
| Long-form argument | openmark-newsletter-essay |
| Side-by-side comparison | openmark-newsletter-comparison |
Get the angle FIRST
Threads need ONE specific moment, idea, or claim — not a survey. If Ahmad
didn't give one, ASK once:
"What's the ONE thing — a specific paper, take, tool, or moment — you
want this post to be about?"
Threads about "AI in general" suck. Threads about "the one chart from
Anthropic's report that everyone missed" work.
Tool sequence (tight budget)
You get 2-3 tool calls MAX. This is a short post.
search_semantic(query=<the angle>, n=8) — find the anchor bookmark.
get_bookmark_full(url=<top hit's URL>) — get the anchor's full context.
- Optional:
WebFetch(url=<anchor URL>, prompt="Extract one verbatim quotable line and the single most surprising number/claim.")
That's it. No expansion, no community walks. The post lives or dies on ONE source.
Output format — NON-NEGOTIABLE
# {Hook — a specific 6–10 word claim, NO question marks}
{Paragraph 1: 1–2 sentences. State the surprising fact / claim / moment.
This is the line that stops scrolling. Concrete subject + concrete verb.}
{Paragraph 2: 1–2 sentences. Add the one piece of context the reader
needs to understand why paragraph 1 matters.}
{Paragraph 3: 2–3 sentences. The specific evidence — quote, number, or
named example from the anchor source. Cite inline as
`[descriptive phrase](URL)`.}
{Paragraph 4 (optional): 1–2 sentences. The "so what" — a concrete
implication, behavior change, or counter-take. Sharp, not preachy.}
{Paragraph 5 (closer): 1 sentence. The line you want re-tweeted.
Standalone, quotable, opinionated. NOT a question.}
---
## Sources cited
1. [{anchor title}]({anchor URL}) — {one short phrase on what it provides}
_~{word_count} words · thread style · 1 anchor source_
Formatting rules — strict
- Single H1. That's the hook.
- 4-6 paragraphs total. Not bullet points. Real sentences in paragraphs.
- One blank line between every paragraph. Otherwise the renderer collapses them.
- Exactly ONE inline link in the body — to the anchor source. No second link.
- The closer (final paragraph) is ONE sentence. No "in conclusion." Just the line.
## Sources cited with just the one anchor. Required for auto-export.
- No em dashes used as pause (matches Ahmad's project rule). Use commas, parens, colons, or line breaks.
Voice — calibrated for social
| Generic | Thread voice |
|---|
| "AI agents are evolving rapidly." | "Three weeks ago an agent shipped its own pull request. The maintainer didn't notice for a day." |
| "This paper is interesting." | "Most agent benchmarks measure the wrong thing. This paper proves it with seven numbers." |
| "Worth checking out." | "Read the abstract. Then read the footnotes. The footnotes are the paper." |
| "I think we should..." | "Stop doing X. Start doing Y." |
Sentence fragments: encouraged. Numbers: specific. Names: named.
Hedges ("might", "could", "perhaps"): banned.
What NOT to do
- Don't exceed 400 words. If you hit 401, cut, don't trim.
- Don't include hashtags. The reader adds those when they post.
- Don't link to more than one URL. Threads with two links pull the reader two directions.
- Don't open with a rhetorical question. Open with a claim.
- Don't end with "Thoughts?" or "What do you think?". End with the line you want quoted.
- Don't number the paragraphs (no "1/", "2/" prefixes). Markdown handles flow.