momentum-check

name: momentum-check description: Computes deterministic momentum, overbought signals, and trend-context flags per ticker — consecutive up-day streak, % change over 5/10/30/90d, distance from 50DMA/150DMA/200DMA/52w-high, RSI(14), 150DMA and 200DMA price-MA crossovers (Weinstein stage-2 setups), and classic 50/200 golden/death crosses. Returns a streak_flag ("normal" | "extended") used to gate BUY recommendations, plus a trend_flags array that is descriptive context for the analyst (never auto-triggers BUY or SELL). Reads yfinance cache via scripts.lib.data — no LLM, no WebSearch.

momentum-check

When to use

• Before the opportunity-scanner LLM lists, to attach exact momentum metrics to each candidate ticker.
• Before any specific-decision BUY recommendation, to verify the name isn't momentum-extended.
• Manual lookup: "is NVDA overbought?" / "how many days has X gone up in a row?"

Why it exists

Replaces the WebSearch-based rally_pct_30d and price_vs_200dma estimates the scanner used to make. Those numbers were LLM guesses from search snippets; these numbers come from cached yfinance closes.

Input

# single ticker
echo '{"ticker":"NVDA"}' | python3 .claude/skills/momentum-check/scripts/momentum.py

# batch (recommended for scanner candidate bundles)
echo '{"tickers":["NVDA","AVGO","META","IWM","TEVA.TA"]}' | python3 .claude/skills/momentum-check/scripts/momentum.py

Output

{
  "data_as_of": "2026-04-29",
  "results": {
    "NVDA": {
      "last_close": 201.68,
      "as_of": "2026-04-28",
      "consecutive_up_days": 4,
      "consecutive_down_days": 0,
      "pct_change_5d": 0.038,
      "pct_change_10d": 0.071,
      "pct_change_30d": 0.272,
      "pct_change_90d": 0.41,
      "pct_vs_50dma": 0.084,
      "pct_vs_150dma": 0.121,
      "pct_vs_200dma": 0.156,
      "days_above_50dma": 18,
      "days_above_150dma": 42,
      "days_above_200dma": 87,
      "trend_flags": [],
      "dist_from_52w_high_pct": -0.012,
      "rsi_14": 68.4,
      "streak_flag": "normal",
      "flags": [],
      "tier": "normal"
    }
  },
  "warnings": []
}

streak_flag rules

streak_flag = "extended" when ANY of:

• consecutive_up_days >= 7
• rsi_14 >= 75
• pct_vs_200dma >= 0.30 (more than 30% above 200DMA)
• pct_change_30d >= 0.25 AND dist_from_52w_high_pct >= -0.02 (rallied hard AND at ATH)

When extended, the flags array names which condition tripped (e.g. ["rsi_overbought","far_above_200dma"]).

trend_flags (descriptive context — NOT actions)

The trend_flags array surfaces moving-average regime changes for the analyst's situational awareness. They are descriptive: a golden_150_cross is bullish context, not an auto-BUY; a death_200_cross is bearish context, not an auto-SELL. The daily-report rule (CLAUDE.md) forbids deriving signals from trend flags alone — the analyst weighs them alongside fundamentals, thesis files, and momentum extension.

Detection rules:

The 20-day prior-run filter on the price-MA crosses prevents whipsaw flagging when price oscillates around the MA line. MA-on-MA crosses don't need that filter — they're inherently slower.

days_above_150dma and days_above_200dma are current consecutive streaks (counted backwards from the latest bar); 0 means the most recent bar closed below the MA.

Notes

• Uses 1y of daily closes from scripts.lib.data.fetch_history. Adjusted closes (auto_adjust=True).
• If fewer than 200 trading days are available (new IPO, illiquid TASE name), 200DMA fields return null and a "insufficient_history" warning is added per ticker.
• TASE tickers must use the .TA suffix.
• Pure Python, no LLM, runs in <2s for 30 tickers from cache.