dragonkid/cex-earn-analyzer

CEX Earn Analyzer

Scan Binance announcements for stablecoin earning activities, verify advertised APY against on-chain reality, and produce a risk-adjusted comparison.

All data fetching uses public APIs — no browser or login required. The bundled scripts/fetch_data.py handles all network calls and returns structured JSON.

Bundled Script

scripts/fetch_data.py provides these subcommands:

SCRIPT="$(dirname "$0")/../scripts/fetch_data.py"   # when called from skill dir
# Or use the absolute path shown by the skill loader

python3 scripts/fetch_data.py announcements [--days N]   # Binance CMS API → JSON list
python3 scripts/fetch_data.py stablecoins                # DefiLlama → JSON list
python3 scripts/fetch_data.py ticker USDCUSDT RLUSDUSDT  # Binance 24h ticker → JSON
python3 scripts/fetch_data.py price WLFIUSDT             # Binance spot price → JSON
python3 scripts/fetch_data.py detail CODE1 CODE2         # defuddle → clean markdown per page
python3 scripts/fetch_data.py parse  CODE1 CODE2         # detail + structured extraction → JSON

All output is JSON to stdout, errors to stderr.

Workflow

fetch_data.py announcements → LLM classify → Exa verify unknowns → fetch_data.py detail → Filter → Verify APY → Output

Step 1: Fetch Announcements

python3 <skill-dir>/scripts/fetch_data.py announcements --days 30

Returns JSON array: [{title, code, date}, ...]

Key notes:

• catalogId=93 = "币安最新活动" category. Only use this category.
• The API returns English titles regardless of page language. Classification must handle English.
• Default range is 30 days. Adjust if the user specifies a different period.

Step 2: LLM Classification (No Keyword Filtering)

Do NOT use keyword-based filtering — it misses new stablecoins and produces false positives (e.g., "Share 300 USDT in rewards" is not a stablecoin earn activity).

Send all titles to the LLM in a single prompt and classify each as:

• stablecoin_earn: The PRIMARY asset the user deposits/holds is a USD-pegged stablecoin, and the activity is earn/savings/holding reward
• non_usd_stablecoin: The asset is a stablecoin pegged to a non-USD fiat currency (e.g., KGST pegged to Kyrgyzstani Som). Include in output separately with a warning — holding these exposes users to FX risk on top of crypto risk, which defeats the purpose of "stable" yield.
• not_relevant: Trading competitions, referral programs, or activities that merely use stablecoins as reward denomination
• uncertain: Cannot determine — coin might be a stablecoin but not sure

What counts as "stablecoin_earn":

• Flexible/locked earn products for a USD-pegged stablecoin (e.g., "USDC Flexible Products", "U活期产品")
• Hold-to-earn airdrops where you hold a USD-pegged stablecoin (e.g., "Hold USD1 to share WLFI")
• Does NOT include: trading competitions with stablecoin prizes, referral bonuses paid in USDT
• Does NOT include by default: non-USD-pegged stablecoins (e.g., KGST/KGS, EURS/EUR). Mention them separately if found.

Note: region restrictions and activity end dates cannot be determined from titles alone — these are filtered in Step 5 after extracting details.

Step 3: Exa Search for Uncertain Coins

For coins classified as "uncertain", search to determine if they are stablecoins:

web_search_exa: "[COIN_NAME] stablecoin OR pegged OR 稳定币"

Reclassify based on results. If still unclear, include in output with a note.

Step 4: Extract Activity Details (Structured)

Use the parse subcommand instead of detail — it fetches the page via defuddle AND automatically extracts structured fields:

python3 <skill-dir>/scripts/fetch_data.py parse CODE1 CODE2 CODE3

Returns JSON array with pre-extracted fields:

[{
  "code": "...", "url": "...",
  "coin": "USDC",           // auto-detected from table
  "type": "Flexible",       // Flexible / Locked / Hold Airdrop
  "advertised_apr": "5.8%", // headline APR
  "tier_limit": 200.0,      // max amount at tier APR (e.g., ≤200 USDC gets 5.8%)
  "tier_apr": "5.8%",       // APR within tier limit
  "base_apr": "0.8%",       // APR above tier limit (Real-Time APR only)
  "personal_limit": 300000000.0,  // per-user max subscription
  "start_time": "2026-04-01 00:00:00 UTC",
  "end_time": "2026-04-30 23:59:59 UTC",
  "is_ended": false,         // auto-compared with current time
  "reward_currency": "USDC", // same coin or different token
  "region_restriction": null, // "Bahrain", "CIS", "GCC", etc. or null
  "new_user_only": false,
  "raw_table_text": "..."   // fallback if auto-extraction missed something
}]

The script handles HTML table parsing, promotion period extraction, and tier limit detection automatically. This avoids the 100KB+ raw markdown problem from detail.

When to fall back to detail: If parse returns null for critical fields (coin, tier_limit) on a specific activity, use detail for that code and parse manually. The raw_table_text field provides a compact summary of the first table for quick LLM review.

Important distinctions:

• U is NOT USDT — "U" is a separate asset on Binance Simple Earn. The announcement explicitly states it is not an abbreviation of USD fiat currency.
• If reward_currency differs from deposit coin (e.g., hold USD1 → earn WLFI), flag this prominently — it introduces token price risk on the reward side.

Step 5: Filter

The parse output already includes is_ended and region_restriction fields. Simply:

• Remove entries where is_ended = true
• Separate entries with region_restriction (report them at the bottom as region-limited)
• Remove entries where new_user_only = true if the user is not a new user (ask if unclear)

Step 6: Verify Real APY

Two categories:

Fixed-rate earn products (e.g., USDC 5.5%, RLUSD 8%): Tiered APY is guaranteed within the tier limit. No verification needed — just note the tier cap.

Variable APY activities (e.g., hold-to-earn airdrops with fixed token pool): The advertised APY is illustrative only. Announcements use hypothetical numbers in examples (e.g., "假设年化收益率20%") — these are NOT promises. Always calculate an estimated range:

6a. Get reward token price:

python3 <skill-dir>/scripts/fetch_data.py price WLFIUSDT

6b. Get stablecoin circulating supply:

python3 <skill-dir>/scripts/fetch_data.py stablecoins

Find the coin by symbol, use market_cap_usd. Assume Binance holds 20-40% of total supply.

6c. Calculate estimated APY range:

annual_reward_value = (pool_per_period * token_price) * periods_per_year
APY = annual_reward_value / total_participating_amount

Conservative: 40% of circulating supply on Binance, 100% participate
Moderate:     40% on Binance, 25% participate
Optimistic:   40% on Binance, 5% participate

Step 7: Risk Assessment

7a. Get market data:

python3 <skill-dir>/scripts/fetch_data.py ticker USDCUSDT RLUSDUSDT UUSDT USD1USDT WLFIUSDT

Returns [{symbol, price, volume_24h_usd}, ...]

For market cap, use the DefiLlama data already fetched in Step 6b.

7b. Assess each coin's risk level:

| Factor | Low | Medium | High | |--------|-----|--------|------| | Issuer track record | Years of operation, major issuer | Backed by known company, < 2 years | New project, unproven | | Regulatory status | Licensed (e.g., NYDFS) | Partial compliance | No clear regulation | | Reserve transparency | Regular third-party audits | Self-reported reserves | Opaque | | 24h trading volume | > $100M | $10M - $100M | < $10M | | Market cap | > $1B | $100M - $1B | < $100M | | Peg stability | Consistent $1.00 | Minor deviations (±0.5%) | Significant deviations | | Reward currency risk | Same coin (USDC→USDC) | N/A | Different volatile token |

Step 8: Output

Tag activities with status icons:

• [NEW] — announced within the last 3 days
• [EXPIRING] — ends within the next 3 days

Output each activity:

## [NEW] [Coin] [Type] — [Advertised APY]

| Field | Detail |
|-------|--------|
| Coin | ... |
| Risk | Low / Medium / High — [one-line reason] |
| Participation | ... |
| APY | ... (with tier breakdown) |
| Time | start → end [EXPIRING: highlight days remaining] |
| Link | announcement URL |

[If variable APY: include estimated range from Step 6]

End with a horizontal comparison table sorted by risk-adjusted return:

| 币种 | APR | 类型 | Tier 限额 | 截止时间 (UTC / 本地) | 风险 | 奖励币种 | 24h Vol | Market Cap | 状态 |

Show both UTC and the user's local timezone. Detect the user's timezone from system environment or conversation context. Example: Apr 30 23:59 UTC / May 1 07:59 UTC+8.

Gotchas from Experience

• All APIs are public — no browser or login needed. The bundled script handles all network calls.
• defuddle for detail pages — returns clean markdown with tables preserved, more reliable than WebFetch's summarization.
• API returns English titles even on the Chinese page. Classification should handle English titles.
• "U" ≠ USDT — always treat as separate assets.
• Advertised APY in examples ≠ guaranteed APY — some announcements use hypothetical numbers in calculation examples. Always flag when APY is variable vs fixed.
• DefiLlama for stablecoin data — stablecoins subcommand returns market cap and circulating supply for all stablecoins.
• Binance ticker API — ticker and price subcommands are public, no API key needed.