X Timeline Digest

Sources:
- For You timeline
- Following timeline
What it does:
1. Fetch recent tweets
2. Filter incrementally (avoid reprocessing)
3. Deduplicate (ID + near-duplicate text)
4. Rank and trim
5. Generate a Chinese digest
6. Output a structured payload
> Delivery (Telegram, email, etc.) is NOT handled here.
> Upstream OpenClaw workflows decide how to notify users.
---
## Configuration
All config is read from: skills.entries["x-timeline-digest"].config
### Config fields
| Name | Type | Default | Description |
|----|----|----|----|
| intervalHours | number | 6 | Interval window in hours |
| fetchLimitForYou | number | 100 | Tweets fetched from For You |
| fetchLimitFollowing | number | 60 | Tweets fetched from Following |
| maxItemsPerDigest | number | 25 | Max tweets in one digest |
| similarityThreshold | number | 0.9 | Near-duplicate similarity threshold |
| statePath | string | ~/.openclaw/state/x-timeline-digest.json | State file path |
---
## Dependencies
- bird must be installed and available in PATH
- bird must already be authenticated (cookie login)
- Read-only usage

## Usage

### 1. Basic (Raw JSON)
Run the digest generator to get a clean, deduplicated JSON payload:
```bash
node skills/x-timeline-digest/digest.js
```

### 2. Intelligent Digest (Recommended)
To generate the "Smart Brief" (Categorized, Summarized, Denoised):
1. Run the script: `node skills/x-timeline-digest/digest.js > digest.json`
2. Read the prompt template: `read skills/x-timeline-digest/PROMPT.md`
3. Send the prompt to your LLM, injecting the content of `digest.json` where `{{JSON_DATA}}` is.

*Note: The script automatically applies heuristic filtering (removes "gm", ads, short spam) before outputting JSON.*

## Bird Commands Used
For You timeline:
bird home -n <N> --json
Following timeline:
bird home --following -n <N> --json
---
## State Management
State is persisted to statePath.
### State structure
{
"lastRunAt": "2026-02-01T00:00:00+08:00",
"sentTweetIds": {
"123456789": "2026-02-01T00:00:00+08:00"
}
}
### Rules
- Tweets already in sentTweetIds must not be included again
- After a successful run:
- Update lastRunAt
- Add pushed tweet IDs to sentTweetIds
- Keep IDs for at least 30 days
---
## Processing Pipeline
1. Fetch from For You and Following
2. Incremental filter using lastRunAt
3. Hard deduplication by tweet id
4. Near-duplicate merge using text similarity
5. Rank and trim to maxItemsPerDigest
6. **Generate a Categorized Chinese Digest** (via PROMPT.md + LLM)
   - Categories: 🤖 AI & Tech, 💰 Crypto & Markets, 💡 Insights, 🗞️ Other
   - Language: Simplified Chinese
   - Format: [Author](URL): Summary
   - Denoising: Remove ads and low-value content
---
## Output
The skill returns one JSON object:
{
"window": {
"start": "2026-02-01T00:00:00+08:00",
"end": "2026-02-01T06:00:00+08:00",
"intervalHours": 6
},
"counts": {
"forYouFetched": 100,
"followingFetched": 60,
"afterIncremental": 34,
"afterDedup": 26,
"final": 20
},
"digestText": "中文摘要内容",
"items": [
{
"id": "123456",
"author": "@handle",
"createdAt": "2026-02-01T02:15:00+08:00",
"text": "tweet text",
"url": "https://x.com/handle/status/123456",
"sources": ["following"]
}
]
}