AgentSkillsCN

tweet-crafter

为 X 平台打造优化于 2026 年 1 月算法的推文、线程及长文帖子。 适用于为 X/Twitter 单独创作帖子、将新闻简报内容适配 X 平台、撰写线程钩子,或优化帖子格式(线程 vs 长文)时使用。包含当前算法信号、Premium 功能,以及面向拉丁美洲受众的西班牙语内容模式。

SKILL.md
--- frontmatter
name: tweet-crafter
description: |
  Craft tweets, threads, and long-form X posts optimized for the January 2026 algorithm.
  Use when creating individual posts for X/Twitter, adapting newsletter content for X,
  writing thread hooks, or optimizing post format (thread vs long-form). Includes current
  algorithm signals, Premium features, and Spanish-language patterns for LATAM audiences.

Tweet Crafter Skill

Algorithm Signals (January 2026)

SignalWeightNotes
First 30 min engagementCriticalMake or break for distribution
RepliesVery High2x value of retweets
RepostsHighNetwork expansion
QuotesHighConversation starter, shown to both audiences
LikesMediumWeakest engagement signal
Premium statusModifier4x in-network, 2x out-of-network
VideoVery High5x engagement vs text (0.42% vs 0.08%)
Dwell timeHighTime spent reading > quick interactions
External linksPenaltyZero median engagement for non-Premium (March 2026)

Key insight: Focus on prompting replies and conversations, not likes. The algorithm heavily rewards content that generates discussion.

2026 Platform Changes

  • Video content dominates X (~80% of user sessions)
  • External links severely penalized for non-Premium accounts
  • Unverified accounts need ~10x more engagement for same reach
  • Threads now underperform long-form articles

Format Selection

code
Long-form (preferred): Tutorials, essays, deep analysis, "X cosas que aprendí", narratives
Thread: Only when content naturally breaks into discrete numbered points

Performance benchmark:
- Long-form articles: Higher engagement, better reach, preferred by algorithm
- Threads: Lower performance, use sparingly for variety

Default to long-form articles. Threads are no longer the go-to format for educational content. Long-form articles consistently outperform threads on reach and engagement.

Link Posts (NEW — October 2025)

Links are no longer penalized. New in-app browser keeps engagement buttons visible.

Requirements for link posts:

  • Write a compelling standalone caption (not just "new post")
  • Include interesting description or insight
  • Add image when possible (150%+ engagement boost)

Per Elon: "posting a link with almost no description will get weak distribution"

Content Templates

Single Tweet (max 280 chars)

code
[Hook/insight/question]

[Optional: 1-2 supporting lines]

[Optional: 1 hashtag max]

Thread Format (Use Sparingly)

code
1/🧵 [Standalone hook that works even without thread]

2/ [First point — strongest insight]

3-N/ [Supporting points]

N/ [CTA: follow, RT first tweet, link to newsletter]

Thread rules:

  • 5-15 tweets optimal
  • Self-reply immediately after posting (signals thread to algorithm)
  • Hook must work standalone — most people only see tweet 1
  • Consider converting to long-form article instead — better performance

Long-Form Post (Premium)

code
[First 280 chars: Hook that appears in timeline — CRITICAL]

[Body: Up to 25,000 chars with bold/italic formatting]

[CTA at end]

Only first 280 characters show in timeline. Front-load the hook.

Media Guidelines

TypeEngagement Multiplier
Video (native, <2:20)10x
Images2-3x
PollsHigh engagement
Plain textBaseline

Video specs: Native upload, under 2:20, no external watermarks.

Hook Formulas

The first line determines everything. Use these proven patterns:

Curiosity:

  • "I used to think [common belief]..."
  • "Here's what nobody tells you about [topic]..."
  • "Everyone's doing [X] wrong. Here's why:"

Value-Forward:

  • "5 ways to [achieve result]:"
  • "The framework I use for [task]:"
  • "[Number] tools that changed how I [outcome]:"

Story:

  • "Last week, something unexpected happened..."
  • "3 years ago, I made a decision that..."
  • "I almost quit [X]. Then I discovered..."

Data/Surprise:

  • "[Surprising statistic] — here's why it matters."
  • "I analyzed [X] and found something interesting:"

Spanish Hooks:

  • "[Número] cosas que aprendí sobre [tema]:"
  • "El error más común en [tema]:"
  • "Nadie te dice esto sobre [tema]:"

Pre-Publish Checklist

Before posting, verify:

  • Hook is in first 280 characters (critical for timeline preview)
  • Content invites replies (questions, opinions, hot takes)
  • 0-2 hashtags maximum
  • Media attached if applicable (video = 5x engagement)
  • No external links unless Premium (or put link in reply)
  • Would I engage with this if I saw it?
  • Posted during optimal hours (9-11 AM or 7-9 PM PST)

Spanish-Language Patterns

Thread Templates

Tutorial Thread:

code
1/🧵 Cómo [accomplish task] en [technology]:

Una guía paso a paso 👇

2/ Paso 1: [Action]
[Explanation]

...

N/ Eso es todo.

Guarda este hilo para cuando lo necesites.
Sígueme @tacosdedatos para más tutoriales en español.

Insights Thread:

code
1/🧵 [Number] cosas que aprendí sobre [topic] después de [experience]:

Un hilo corto 👇

2/ [Insight #1]
[Brief explanation]

...

N/ Si te fue útil:
• Sígueme para más contenido de [topic]
• RT el primer tweet para que llegue a más personas

Newsletter Promotion

code
[Insight or hook from the newsletter]

Esto y más en el nuevo tacosdedatos:
[Link]

[Optional: What else is in the issue]

Hashtag Strategy

  • Maximum 1-2 hashtags (X algorithm doesn't favor heavy usage)
  • Spanish hashtags for discoverability: #DataScience #Python #DatosAbiertos
  • Skip hashtags entirely for conversational posts

Timing

  • Best times: 9-11 AM, 7-9 PM (PST)
  • Best days: Tuesday-Thursday
  • Post when followers are online (check analytics)

Output Format

When crafting tweets, provide:

markdown
## Tweet/Thread: [Topic]

### Content
[Full post content]

### Format
- **Type**: Single tweet / Thread (X tweets) / Long-form
- **Character count**: X/280 (or full length for long-form)
- **Media needed**: [Image description / Video spec / None]

### Algorithm Notes
- Hook strength: [Strong/Medium/Weak]
- Suggested posting time: [Day, Time]

References

For deep algorithm research and source links, see references/REFERENCES.md.