Binance Square Signal Agent

End-to-end trading signal pipeline built on the thesis that market makers seed narratives on Binance Square before / during accumulation. Square is the largest unblocked Chinese crypto info source — tracking narrative flow there is the sentiment equivalent of tracking on-chain fund flow.

First-Time Setup

The skill ships with two Node.js scripts that need puppeteer-core installed once. After ClawHub installs this skill to ~/.claude/skills/binance-square/, run:

CODEBLOCK0

Then verify Chrome/Chromium is installed (auto-detected on Win/Mac/Linux). Override with CHROME_PATH env var if needed.

Optional Configuration (env vars)

Modes (dispatch on $ARGUMENTS)

Pipeline: scan and scan:tg

Step 1 — Scrape

CODEBLOCK1

The scraper writes a square-YYYY-MM-DD-HHmm.json file in its own directory by default. Override with --out PATH. Read the JSON output. Key fields per coin in coinRanking:

• - mentions, botPct, INLINECODE23
• INLINECODE24 (BULLISH / BEARISH / NEUTRAL), sentiment.score, sentiment.bullish, INLINECODE27
• INLINECODE28, INLINECODE29

INLINECODE30 contains the same breakdown for each drilled hashtag page (larger sample, bot-dense — this is where coordinated narrative pushes are visible).

Step 2 — On-chain confirmation (skip if COINGLASS_BASE not set)

For each non-baseline coin (not BTC/ETH/BNB/SOL/XRP/DOGE) with 2+ feed mentions:

CODEBLOCK2

Extract:

• - OI: total.chg24h (24h OI change %)
• Funding: average rate across major exchanges (Binance / Bybit / OKX)
• Liquidation: long24h vs short24h USD per coin
• Price 24h: from gainers (Step 3) or derive from OI deltas

The user must provide their own COINGLASS_BASE URL — either the official Coinglass API with their key, or their own proxy. Without this, direction judgment falls back to Square sentiment + price action only.

Step 3 — Gainers cross-reference

WebFetch https://www.binance.com/zh-CN/markets/coinInfo — extract 领涨榜 (top gainers) with 24h% change.

Step 4 — Direction judgment

Combine the four signal sources for each candidate coin:

CODEBLOCK3

Direction call: LONG ✅ / SHORT 🔻 / AVOID ⚠️ / WATCH 👀

• - WATCH for OI divergence with unclear direction
• INLINECODE39 for bot-dominated noise or event-driven moves (FUD / hack / legal)

Step 5 — Save full report

Write detailed markdown report to ~/.claude/skills/binance-square/reports/signal-YYYY-MM-DD-HHmm.md (create the reports/ dir if needed). Include:

• - Candidates table (coin, posts, bot%, sentiment, OI, funding, liq ratio, direction)
• Per-coin direction rationale
• Drill bot breakdown
• Gainers cross-ref
• Risk notes

Step 6 — Telegram push (only if scan:tg and TG env vars set)

Write condensed summary (<4000 chars) to a temp file:

CODEBLOCK4

If TG_BOT_TOKEN or TG_CHAT_ID env var is missing, skip this step and return the message text in the response instead.

Mode: coin:TICKER

Deep-drill a specific coin's topic page (~200 posts vs ~20 on main feed):

CODEBLOCK5

Then run Step 2 (on-chain) + Step 4 (direction) for just this coin. Report:

CODEBLOCK6

Mode: read:URL or read:KEYWORD

URL form (read:https://www.binance.com/...)

WebFetch the article. Extract: title, author, publish time, content, engagement, coin mentions, hashtags. Flag author if username matches Square-Creator-xxx → potential bot.

Keyword form (read:RAVE 爆仓)

WebSearch site:binance.com/square KEYWORD, then WebFetch top 3-5 results. For each, extract summary + sentiment + author type. Aggregate: overall sentiment, bot %, narrative theme.

Bot Detection (built into scraper)

Layer 1 (username pattern):

• - Display name matches /^Square-Creator-[a-f0-9]+$/ → BOTSUSPECT (never customized profile)
• Otherwise → tentatively LIKELYHUMAN

Layer 2 (behavioral, post-hoc):

• - Default username (square-creator-* profile ID) + 3+ posts in batch + avg views < 200 → BOTSUSPECT
• Any author with 5+ posts + avg views < 100 → BOTSUSPECT (catches custom-named bots)

Sentiment scoring only counts non-bot posts to avoid contamination.

The drill mode reveals bot rings — accounts that appear across multiple coin hashtags with high post counts and low engagement = signature of coordinated narrative promotion.

Output Contract

Return to main agent (concise — full detail lives in the saved report):

1. 1. Top 3 actionable signals with direction (LONG/SHORT/AVOID) + one-sentence rationale each
2. Path to full saved report
3. If scan:tg: confirmation of TG send with message_id, OR the message text + note that env vars weren't configured
4. Critical anomalies: new coin trending, extreme funding rate, bot farm coordination

Disclaimer

This tool finds patterns. It is not financial advice. Trading decisions and risk management are entirely the user's responsibility. Past pattern correlations do not predict future moves.