home / skills / questnova502 / claude-skills-sync / weibo-autopilot

weibo-autopilot skill

/skills/weibo-autopilot

This skill automates Weibo browsing and reposting, learns user preferences, and comments thoughtfully to enhance engagement.

npx playbooks add skill questnova502/claude-skills-sync --skill weibo-autopilot

Review the files below or copy the command above to add this skill to your agents.

Files (11)
SKILL.md
4.0 KB
---
name: liubin-weibo-autopilot
description: Autonomous Weibo browsing and reposting agent. Learns user preferences, browses feeds, and reposts interesting content with thoughtful commentary.
---

# Weibo Autopilot (微博自动驾驶)

An autonomous agent that browses Weibo feeds, identifies interesting content based on learned user preferences, and reposts with thoughtful commentary.

## Script Directory

**Important**: All scripts are located in the `scripts/` subdirectory of this skill.

**Agent Execution Instructions**:
1. Determine this SKILL.md file's directory path as `SKILL_DIR`
2. Script path = `${SKILL_DIR}/scripts/<script-name>.ts`
3. Replace all `${SKILL_DIR}` in this document with the actual path

**Script Reference**:
| Script | Purpose |
|--------|---------|
| `scripts/learn-preferences.ts` | Learn user posting/commenting habits |
| `scripts/browse-feed.ts` | Browse Weibo feeds and extract content |
| `scripts/repost.ts` | Repost content with commentary |
| `scripts/autopilot.ts` | Main controller for continuous operation |
| `scripts/weibo-cdp.ts` | Shared Chrome CDP utilities |

## Prerequisites

- Google Chrome or Chromium installed
- `bun` installed (for running scripts)
- Logged in to Weibo (will prompt on first run)

## Data Files

User preferences and learning data are stored in `data/` subdirectory:
- `data/user-profile.json` - Learned user preferences and posting style
- `data/repost-history.json` - History of reposts to avoid duplicates

---

## Commands

### 1. Learn User Preferences

Learn from user's posting history and comment style. Run periodically (e.g., weekly) to update preferences.

```bash
# Learn from user's Weibo profile and comment history
npx -y bun ${SKILL_DIR}/scripts/learn-preferences.ts

# Force refresh (ignore cached data)
npx -y bun ${SKILL_DIR}/scripts/learn-preferences.ts --refresh
```

**Data Sources**:
- User profile: https://weibo.com/u/1043848755
- Comment outbox: https://weibo.com/comment/outbox

---

### 2. Start Autopilot Mode

Start continuous browsing and reposting. Runs in background, reposts approximately every 10 minutes.

```bash
# Start autopilot (default: browse all feeds)
npx -y bun ${SKILL_DIR}/scripts/autopilot.ts

# Browse specific feed group
npx -y bun ${SKILL_DIR}/scripts/autopilot.ts --group "国际新闻/军事"
npx -y bun ${SKILL_DIR}/scripts/autopilot.ts --group "足球"

# Set repost interval (default: 10 minutes with ±3 min variance)
npx -y bun ${SKILL_DIR}/scripts/autopilot.ts --interval 15

# Dry run (browse but don't actually repost)
npx -y bun ${SKILL_DIR}/scripts/autopilot.ts --dry-run
```

**Parameters**:
| Parameter | Description |
|-----------|-------------|
| `--group <name>` | Browse specific feed group (optional) |
| `--interval <min>` | Average minutes between reposts (default: 10) |
| `--dry-run` | Browse and analyze but don't repost |

---

### 3. Manual Repost

Manually repost a specific Weibo with commentary.

```bash
# Repost with auto-generated commentary
npx -y bun ${SKILL_DIR}/scripts/repost.ts <weibo-url>

# Repost with custom commentary
npx -y bun ${SKILL_DIR}/scripts/repost.ts <weibo-url> --comment "Your commentary here"
```

---

## Behavior Rules

### Content Selection
- Analyzes text, images (via vision), video descriptions, and comments
- Matches content against learned user interests
- Avoids content that might be controversial or offensive
- Skips already reposted content

### Commentary Generation
- Thoughtful, non-controversial observations
- Respects original poster and other users
- Matches user's learned commenting style
- Always includes AI signature: ` ⸢ᴬᴵ ᵖᵒˢᵗᵉᵈ ᵛⁱᵃ ᶜˡᵃᵘᵈᵉ ᶜᵒᵈᵉ⸥`

### Safety Guidelines
- Never comments, only reposts
- Avoids politically sensitive content
- Respects all users and groups
- Fully understands content before reposting

---

## Notes

- First run requires manual Weibo login (session is saved)
- Learning preferences should be run before first autopilot session
- Browser runs in background during autopilot mode
- Logs are written to stdout for monitoring

Overview

This skill is an autonomous Weibo browsing and reposting agent that learns a user's preferences, scans feeds, and reposts interesting posts with thoughtful commentary. It runs headless in a Chromium browser, maintains a repost history to avoid duplicates, and tailors commentary to the user's learned style while including an AI signature. It is designed to operate continuously in autopilot mode or be invoked manually for single reposts.

How this skill works

The agent first learns the user's posting and commenting habits from their profile and comment history, storing a compact preference profile. It then browses selected Weibo feeds, analyzes text, images, video descriptions, and comments, scores content against the learned preferences, and selects candidates that meet safety rules. When reposting, it generates a respectful, non-controversial commentary that matches the user's voice, appends the required AI signature, and records the repost in history to prevent repeats.

When to use it

  • Automate regular sharing of content that matches your interests without manual browsing.
  • Keep a consistent repost cadence while you focus on other tasks (use interval controls).
  • Seed your timeline with curated, thoughtful commentary in your learned voice.
  • Dry-run analysis to audit what would be reposted before enabling actual posts.
  • Manually repost a specific Weibo with autogenerated or custom commentary.

Best practices

  • Run the learning script periodically (e.g., weekly) so the preference profile stays current.
  • Start with a dry run to validate selection and commentary before enabling autopilot reposts.
  • Set reasonable repost intervals and variance to mimic natural behavior (default 10 ±3 minutes).
  • Review repost-history regularly to ensure no unwanted duplicates or false positives.
  • Restrict feed groups to narrow topics if you want focused reposting behavior.

Example use cases

  • A sports fan automates reposting of match highlights and commentary aligned with their tone.
  • A news curator continuously surfaces relevant international updates while avoiding sensitive topics.
  • An influencer maintains activity by reposting selected posts with personalized, respectful remarks.
  • A researcher uses dry-run mode to analyze what content the agent would prioritize for later tuning.

FAQ

Does the agent post immediately after learning preferences?

No. Learning builds or refreshes the preference profile; autopilot reposting runs separately and should be started explicitly.

Can I prevent reposts to specific topics or accounts?

Yes. Configure feed groups and use conservative preference filtering or run targeted groups only to avoid unwanted sources.