Twitter Hand

openfang hand activate twitter

openfang hand config twitter \
  --set content_style="witty" \
  --set content_topics="AI agents, automation, productivity" \
  --set brand_voice="Technical founder who makes AI accessible with humor" \
  --set post_frequency="3_daily" \
  --set approval_mode="true"

Style: Witty
Topics: AI agents, automation, productivity
Frequency: 3 tweets per day (8 AM, 12 PM, 5 PM)
Mode: Approval (queue for review)

> Generate 10 tweets about AI agent trends

curl -H "Authorization: Bearer $TWITTER_BEARER_TOKEN" \
  "https://api.twitter.com/2/users/me"

web_search("AI agents trending today")
web_search("productivity latest news")
web_search("automation viral tweets") // for format inspiration only

curl -H "Authorization: Bearer $TWITTER_BEARER_TOKEN" \
  "https://api.twitter.com/2/tweets/search/recent?query=AI%20agents&max_results=10&tweet.fields=public_metrics"

// Professional style
"AI agents represent a fundamental shift in how we interact with software systems."

// Casual style
"ngl ai agents are kinda wild when you see them in action 🤖"

// Witty style
"AI agents are like interns who never sleep, complain, or ask for equity. Perfect."

// Educational style
"Let's break down how AI agents work:
1. LLM generates plan
2. Tools execute actions
3. Results feed back
4. Repeat until done"

// Provocative style
"Hot take: If your 'AI agent' can't handle failures, it's not an agent — it's a script."

// Inspirational style
"Imagine a world where every repetitive task runs itself. We're building that future. ✨"

[
  {
    "id": "q_001",
    "content": "Unpopular opinion: Most 'AI agents' are just chatbots with extra steps.\n\nReal agents:\n• Plan multi-step workflows\n• Use tools autonomously\n• Recover from failures\n
Chatbots with function calling aren't agents. They're assistants.\n\nThe bar is higher.",
    "type": "hot_take",
    "created": "2026-03-06T10:30:00Z",
    "status": "pending"
  },
  {
    "id": "q_002",
    "content": "Build your first AI agent in 5 lines of Python:\n\n```python\nfrom anthropic import Anthropic\nclient = Anthropic()\nresponse = client.messages.create(\n  model='claude-3-5-sonnet',\n  tools=[calculator, web_search],\n  messages=[{'role': 'user', 'content': 'Research AI trends'}]\n)\n```\n\nThat's it. Claude handles the rest.",
    "type": "tip",
    "created": "2026-03-06T10:31:00Z",
    "status": "pending"
  }
]

# Twitter Queue — 10 tweets ready

## Tweet 1 (Hot take)
Unpopular opinion: Most 'AI agents' are just chatbots with extra steps.

Real agents:
• Plan multi-step workflows
• Use tools autonomously
• Recover from failures

Chatbots with function calling aren't agents. They're assistants.

The bar is higher.

---

## Tweet 2 (Tip)
Build your first AI agent in 5 lines of Python:

```python
from anthropic import Anthropic
...

To approve and post:
1. Review `twitter_queue_preview.md`
2. Edit if needed
3. Tell Twitter Hand: "Post the queue" or "Post tweet #1"

### 5. Posting (Manual Approval or Auto)

If approved:

**Post a tweet:**
```bash
curl -X POST "https://api.twitter.com/2/tweets" \
  -H "Authorization: Bearer $TWITTER_BEARER_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"text": "tweet content here"}'

# Tweet 1
curl -X POST "https://api.twitter.com/2/tweets" \
  -d '{"text": "Thread hook tweet"}' \
  -o response1.json

# Extract tweet_id from response1.json
FIRST_ID=$(jq -r '.data.id' response1.json)

# Tweet 2 (reply to tweet 1)
curl -X POST "https://api.twitter.com/2/tweets" \
  -d '{"text": "Thread continuation", "reply": {"in_reply_to_tweet_id": "'$FIRST_ID'"}}' \
  -o response2.json

# Repeat for remaining tweets in thread

{
  "id": "1234567890",
  "text": "...",
  "posted_at": "2026-03-06T12:00:00Z",
  "type": "hot_take",
  "metrics": null // updated later
}

curl -H "Authorization: Bearer $TWITTER_BEARER_TOKEN" \
  "https://api.twitter.com/2/users/$USER_ID/mentions?max_results=10&tweet.fields=created_at"

// Mention: "@you What's the best agent framework?"
// Generated reply (matching your style):
"Great question! For production, I'd go with LangGraph (flexibility) or Claude's native tool use (simplicity). Depends if you need complex workflows or straightforward tasks."

curl -X POST "https://api.twitter.com/2/users/$USER_ID/likes" \
  -H "Authorization: Bearer $TWITTER_BEARER_TOKEN" \
  -d '{"tweet_id": "1234567890"}'

curl -H "Authorization: Bearer $TWITTER_BEARER_TOKEN" \
  "https://api.twitter.com/2/tweets?ids=ID1,ID2,ID3&tweet.fields=public_metrics"

{
  "data": [
    {
      "id": "1234567890",
      "text": "...",
      "public_metrics": {
        "retweet_count": 12,
        "reply_count": 8,
        "like_count": 45,
        "quote_count": 3,
        "bookmark_count": 7,
        "impression_count": 2341
      }
    }
  ]
}

Generate 5 tweets about [breaking news topic]

Write a 10-tweet thread explaining AI agent architectures for developers

Reply to this mention with a helpful response: [tweet URL or text]

Analyze my tweet performance from the last month. What's working?

cat twitter_queue.json | jq -r '.[] | "\(.created): \(.content)"'