x-algo-skills

内容来源：README.md（说明文档） · 原始地址 · 查看安装指南

原始内容

# X Algorithm Skills for AI Coding Agents

AI-powered skills that help you understand and leverage the X (Twitter) recommendation algorithm. Add these to your coding agent (Claude Code, Cursor, etc.) to analyze posts, understand ranking, and optimize your X presence.

## What You Can Do

Ask your AI assistant questions like:

- _"Why did my post get low engagement?"_
- _"What engagement signals does X track?"_
- _"How does the X algorithm rank posts?"_
- _"Why was my post filtered from the feed?"_
- _"Analyze this post for X algorithm optimization"_

## Skills Included

| Skill          | Command              | Use Case                                      |
| -------------- | -------------------- | --------------------------------------------- |
| **Engagement** | `/x-algo-engagement` | Understand the 18 engagement signals X tracks |
| **Scoring**    | `/x-algo-scoring`    | Learn how posts are scored and weighted       |
| **Filters**    | `/x-algo-filters`    | Know why posts get filtered out               |
| **Pipeline**   | `/x-algo-pipeline`   | Full overview of the recommendation system    |
| **ML Model**   | `/x-algo-ml`         | Deep dive into the Phoenix transformer model  |

## Installation

### Claude Code

Copy skills to your project or global skills directory:

```bash
# Project-scoped (recommended)
cp -r x-algo-* /path/to/your/project/.claude/skills/

# Or global
cp -r x-algo-* ~/.claude/skills/
```

### Cursor / Other Agents

Add the skill files to your agent's context or skills directory. Each skill is a standalone `SKILL.md` file with YAML frontmatter.

## Example Interactions

### Analyze a Post

```
You: Analyze this post for X algorithm optimization:
"Just shipped a new feature! Check it out at [link]"

AI: Based on the X algorithm scoring system:
- Low predicted engagement signals (no question, no visual)
- Link posts typically get lower reach due to X preferring native content
- Suggestions: Add an image, ask a question, remove external link...
```

### Understand Filtering

```
You: Why might my post not appear in followers' feeds?

AI: The X algorithm has 12 filters. Your post could be filtered by:
- AgeFilter: Post is too old (>24-48h typically)
- DedupConversationFilter: Another post in the thread scored higher
- VFFilter: Safety/spam detection triggered
- ...
```

### Learn the Algorithm

```
You: How does X decide which posts to show me?

AI: X uses an 8-stage pipeline:
1. Query Hydration - Load your preferences
2. Sources - Thunder (following) + Phoenix (For You)
3. Candidate Hydration - Fetch post data
4. Pre-Score Filtering - Remove ineligible posts
5. Scoring - ML predictions → weighted scores
6. Selection - Top K by score
7. Post-Score Filtering - Dedup, seen posts
8. Side Effects - Logging
```

## Key Insights from the Algorithm

### Engagement Weights

The algorithm predicts probability of 18 actions and combines them:

**Positive signals** (increase score):

- Favorite (like)
- Reply
- Retweet
- Quote tweet
- Follow author
- Share (DM, copy link)
- Video quality view
- Dwell time

**Negative signals** (decrease score):

- Not interested
- Block author
- Mute author
- Report

### What Gets Filtered

Posts are removed if they:

- Are too old (Snowflake ID age check)
- Were already seen (Bloom filter)
- Are from blocked/muted accounts
- Contain muted keywords
- Fail safety checks
- Are duplicates or from same conversation

## Source

These skills are based on the [open-source X recommendation algorithm](https://github.com/twitter/the-algorithm) and internal documentation, adapted for practical use with AI coding agents.

## Contributing

PRs welcome! Areas to improve:

- Add more example queries
- Update weights if X publishes changes
- Add skills for specific use cases (video optimization, thread strategy)

## License

MIT

X Algorithm Skills for AI Coding Agents

AI-powered skills that help you understand and leverage the X (Twitter) recommendation algorithm. Add these to your coding agent (Claude Code, Cursor, etc.) to analyze posts, understand ranking, and optimize your X presence.

What You Can Do

Ask your AI assistant questions like:

"Why did my post get low engagement?"
"What engagement signals does X track?"
"How does the X algorithm rank posts?"
"Why was my post filtered from the feed?"
"Analyze this post for X algorithm optimization"

Skills Included

Skill	Command	Use Case
Engagement	`/x-algo-engagement`	Understand the 18 engagement signals X tracks
Scoring	`/x-algo-scoring`	Learn how posts are scored and weighted
Filters	`/x-algo-filters`	Know why posts get filtered out
Pipeline	`/x-algo-pipeline`	Full overview of the recommendation system
ML Model	`/x-algo-ml`	Deep dive into the Phoenix transformer model

Installation

Claude Code

Copy skills to your project or global skills directory:

# Project-scoped (recommended)
cp -r x-algo-* /path/to/your/project/.claude/skills/

# Or global
cp -r x-algo-* ~/.claude/skills/

Cursor / Other Agents

Add the skill files to your agent's context or skills directory. Each skill is a standalone SKILL.md file with YAML frontmatter.

Example Interactions

Analyze a Post

You: Analyze this post for X algorithm optimization:
"Just shipped a new feature! Check it out at [link]"

AI: Based on the X algorithm scoring system:
- Low predicted engagement signals (no question, no visual)
- Link posts typically get lower reach due to X preferring native content
- Suggestions: Add an image, ask a question, remove external link...

Understand Filtering

You: Why might my post not appear in followers' feeds?

AI: The X algorithm has 12 filters. Your post could be filtered by:
- AgeFilter: Post is too old (>24-48h typically)
- DedupConversationFilter: Another post in the thread scored higher
- VFFilter: Safety/spam detection triggered
- ...

Learn the Algorithm

You: How does X decide which posts to show me?

AI: X uses an 8-stage pipeline:
1. Query Hydration - Load your preferences
2. Sources - Thunder (following) + Phoenix (For You)
3. Candidate Hydration - Fetch post data
4. Pre-Score Filtering - Remove ineligible posts
5. Scoring - ML predictions → weighted scores
6. Selection - Top K by score
7. Post-Score Filtering - Dedup, seen posts
8. Side Effects - Logging

Key Insights from the Algorithm

Engagement Weights

The algorithm predicts probability of 18 actions and combines them:

Positive signals (increase score):

Favorite (like)
Reply
Retweet
Quote tweet
Follow author
Share (DM, copy link)
Video quality view
Dwell time

Negative signals (decrease score):

Not interested
Block author
Mute author
Report

What Gets Filtered

Posts are removed if they:

Are too old (Snowflake ID age check)
Were already seen (Bloom filter)
Are from blocked/muted accounts
Contain muted keywords
Fail safety checks
Are duplicates or from same conversation

Source

These skills are based on the open-source X recommendation algorithm and internal documentation, adapted for practical use with AI coding agents.

Contributing

PRs welcome! Areas to improve:

Add more example queries
Update weights if X publishes changes
Add skills for specific use cases (video optimization, thread strategy)

License

MIT