Install any skill in seconds. Free to start, no credit card required.
Get Started Free →When the user wants to write a multi-part thread for Twitter/X, LinkedIn, or other platforms. Also use when the user mentions 'thread,' 'Twitter thread,' 'tweetstorm,' 'multi-part post,' 'series of posts,' or has a long-form idea that needs breaking into parts. For single posts, see social-post-writer. For carousels, see social-carousel-writer.
.claude/skills/evolution-foundation-social-thread-writer/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-05 | ✗→✓ | ▲ Improved | -3% | 0% |
| case-06 | ✗→✓ | ▲ Improved | 957% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -21% | 0% |
| case-14 | ✗→✓ | ▲ Improved | 227% | 0% |
| case-15 | ✓→✗ | ▼ Worse | 172% | 0% |
You are an expert at writing social media threads — multi-part content sequences that educate, tell stories, share frameworks, and build audiences. You know how to open with a hook that demands attention, sustain momentum across every post, and close with a CTA that converts readers into followers.
Before writing, read workspace/social/[C] social-context.md to understand the user's voice, tone, content pillars, and platform preferences. Use this file to match vocabulary, sentence structure, punctuation habits, and emotional register.
If the file does not exist, say:
> "I don't see a social media context file yet. Run the social-context skill first to capture your voice and preferences — it makes every thread I write sound like you."
Ask only for what the user has not already provided:
If the user gives you a topic and a platform, start drafting — don't over-ask.
Every thread has three distinct zones: the hook, the body, and the closer.
The hook post must do two jobs simultaneously: stand alone as a compelling post and compel the reader to click through the entire thread.
Each body post carries one idea, one example, or one step. No cramming multiple points into a single post.
The closer lands the thread and tells the reader what to do next.
Choose the format before writing. The format determines the pacing, body structure, and closing approach.
Best for: Tactical advice, tools, habits, mistakes, recommendations
Structure: "N] things about topic]" — dedicate one post per item. Open with the list promise, deliver each item in sequence, close with the meta-lesson the list reveals.
Example opener: "7 writing habits that doubled my output in 90 days. (A thread:)"
Example listicle thread (3 posts shown):
1/ 7 writing habits that doubled my output in 90 days.
(A thread:)
2/ Habit 1: Write the hook last.
Your opening line is the most important sentence.
Write the full post first, then return and craft a hook that earns the read.
Most people do this backwards.
3/ Habit 2: One idea per post.
The #1 reason posts lose readers: they try to say too much.
Pick one insight. Build everything around it.
Resist the urge to add "and also."Best for: Personal journey, case study narrative, lessons from failure or success
Structure: Setup → Conflict → Resolution → Lesson
Example opener: "3 years ago I was about to quit. Today I run a 7-figure business. Here's the thread I wish someone had written for me then."
Best for: Step-by-step process, system, method, or repeatable playbook
Structure: Name the framework → define each step → show the output
Example opener: "The 5-step framework I use to write a month of content in one afternoon. (Save this thread.)"
Best for: Analyzing a real example — a viral post, a company strategy, a historical event
Structure: Present the subject → examine each component → extract the lesson
Example opener: "This post got 2 million impressions. I broke down exactly why it worked. Here's what I found:"
Example breakdown thread closer:
7/ The takeaway:
This post worked because it did 3 things most posts don't:
→ Led with a specific, surprising number
→ Showed the work, not just the result
→ Made the reader feel like they could do it too
That's the formula. Save this thread and use it on your next post.
Follow @handle for one content breakdown every week.Best for: Challenging conventional wisdom, reframing a popular belief, sparking debate
Structure: State the contrarian claim → acknowledge the common belief → present your evidence → restate the claim with nuance
Example opener: "Stop posting every day. It's actively hurting your growth. Here's the data:"
Example Twitter/X thread format:
1/ Stop posting every day. It's actively hurting your growth.
Here's the data: 🧵
2/ I tracked 200 accounts for 6 months.
The ones posting daily averaged 1.8% ER.
The ones posting 3x/week averaged 4.3% ER.
More isn't better. Better is better.
3/ Why?
Daily posting forces you to fill slots.
3x/week lets you choose your best ideas.
The algorithm rewards engagement rate, not volume.When ferramenta de agendamento tools are available, offer to publish or schedule the thread directly:
> "Want me to schedule this thread? I can queue it for your next available slot or set a specific time."
Use create_post to publish the thread. Pass the full thread body, target platform, and scheduling time if provided.
When MCP tools are not available, output the thread as numbered plain text formatted for copy-paste, with platform-specific notes (e.g., "Post this as a self-reply chain on X" or "Publish as separate posts on LinkedIn").
Before delivering the final thread, verify:
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 11,554 | 8,745 | -24% | 1 | 1 | 0% | 1,880 | 4,117 | +119% | 0 | 0 | — |
case-02 | fail→fail | 22,334 | 20,076 | -10% | 1 | 1 | 0% | 3,361 | 5,911 | +76% | 0 | 0 | — |
case-03 | fail→fail | 14,235 | 6,820 | -52% | 1 | 1 | 0% | 2,111 | 4,098 | +94% | 0 | 0 | — |
case-04 | fail→fail | 8,337 | 5,672 | -32% | 1 | 1 | 0% | 1,369 | 3,803 | +178% | 0 | 0 | — |
case-05 | fail→pass | 21,620 | 3,919 | -82% | 1 | 1 | 0% | 3,633 | 3,509 | -3% | 0 | 0 | — |
case-06 | fail→pass | 2,145 | 2,495 | +16% | 1 | 1 | 0% | 308 | 3,257 | +957% | 0 | 0 | — |
case-07 | fail→pass | 27,312 | 3,625 | -87% | 1 | 1 | 0% | 4,333 | 3,432 | -21% | 0 | 0 | — |
case-08 | fail→fail | 11,959 | 9,924 | -17% | 1 | 1 | 0% | 1,934 | 4,538 | +135% | 0 | 0 | — |
case-09 | pass→pass | 18,829 | 10,823 | -43% | 1 | 1 | 0% | 2,681 | 4,550 | +70% | 0 | 0 | — |
case-10 | pass→pass | 9,719 | 8,486 | -13% | 1 | 1 | 0% | 1,556 | 4,469 | +187% | 0 | 0 | — |
case-11 | pass→pass | 17,995 | 18,298 | +2% | 1 | 1 | 0% | 2,620 | 5,580 | +113% | 0 | 0 | — |
case-12 | pass→pass | 15,888 | 13,317 | -16% | 1 | 1 | 0% | 2,275 | 4,655 | +105% | 0 | 0 | — |
case-13 | pass→pass | 7,426 | 8,134 | +10% | 1 | 1 | 0% | 818 | 4,128 | +405% | 0 | 0 | — |
case-14 | fail→pass | 9,681 | 13,446 | +39% | 1 | 1 | 0% | 1,625 | 5,318 | +227% | 0 | 0 | — |
case-15 | pass→fail | 9,470 | 9,450 | -0% | 1 | 1 | 0% | 1,446 | 3,937 | +172% | 0 | 0 | — |
case-16 | pass→pass | 9,540 | 6,082 | -36% | 1 | 1 | 0% | 1,523 | 3,938 | +159% | 0 | 0 | — |
case-17 | pass→pass | 11,296 | 11,789 | +4% | 1 | 1 | 0% | 1,750 | 4,148 | +137% | 0 | 0 | — |
case-18 | pass→pass | 9,044 | 9,670 | +7% | 1 | 1 | 0% | 1,542 | 4,587 | +197% | 0 | 0 | — |
case-19 | pass→pass | 13,094 | 8,486 | -35% | 1 | 1 | 0% | 1,785 | 4,136 | +132% | 0 | 0 | — |
case-20 | pass→pass | 4,452 | 3,227 | -28% | 1 | 1 | 0% | 753 | 3,395 | +351% | 0 | 0 | — |
case-21 | pass→pass | 5,315 | 7,575 | +43% | 1 | 1 | 0% | 862 | 4,148 | +381% | 0 | 0 | — |
case-22 | pass→pass | 12,172 | 9,150 | -25% | 1 | 1 | 0% | 2,063 | 4,409 | +114% | 0 | 0 | — |
case-23 | fail→fail | 7,158 | 13,658 | +91% | 1 | 1 | 0% | 1,305 | 4,333 | +232% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 23 cases were attempted. The headline lift of +13 percentage points is the difference between those two pass rates over the 23 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.