stock-deep-dive

Files (2)

• SKILL.md
• dossier.mjs · scripts/

SKILL.md (first lines)

# Stock Deep Dive — everything on one ticker

This is the **single-stock** counterpart to `stock-recap` (which scans the whole
market). Here the user hands you **one ticker** and wants the complete read fast.
It exists so Michael can ask *you* "do you like PLTR down here?" instead of pulling
up six tabs of technicals, options, ownership, and news himself.

It pulls four legs for the ticker and you add a fifth (news):

1. **Technicals** — TD Pro `/api/agent/ticker/:SYM`: RSI, ADX, stoch, MACD, ATR,
   the full EMA/SMA ladder (8/21/55/89/50/200), trend status, EMA stack, buy-zone,
   squeeze ratio, Williams %R, CCI, Bollinger %B, pivots (S2→R2), fib retrace
   (.382/.500/.618), Keltner channel, 90-day range, and bounce state.
2. **Options positioning** — same endpoint: max pain, call wall (resistance), put
   wall (support), expected move to next expiry, IV skew + dealer sentiment,
   put/call OI ratio, and the biggest open-interest strikes (call vs put OI).
3. **Options flow** — TD Pro `/api/agent/unusual-activity?ticker=SYM` over the last
   week: net bullish vs bearish premium, call vs put premium, the largest trades,
   institutional-alpha tier, and repeat-strike conviction.
4. **Institutional ownership** — TickerTrace `/ticker/:SYM`: which funds/ETFs hold
   it and at what weight, plus **recent 13F changes** (adds/trims, including option
   overlays). Read the provider + type — a YieldMax/Roundhill option overlay is
   income mechanics, not a directional conviction bet.
5. **News / catalysts** — *not in the script* (no TD Pro news endpoint). After the
   pull, **you WebSearch** for recent headlines, earnings dates, guidance, analyst
   moves — whatever explains where price is — and fold it into the verdict.

It also **renders a 90-day candlestick chart** (reusing the `stock-recap` skill's
Python venv + `render_chart.py` — no second install).

## How to run it

1. Run the dossier script with the ticker(s). Works from any directory — the
   absolute path resolves the symlink back into the mphinance repo, so the API key
   and the sibling `stock-recap` chart venv are found automatically:

   ```bash
   node "$HOME/.claude/skills/stock-deep-dive/scripts/dossier.mjs" PLTR
   ```

   - One ticker is the norm; pass several (`… dossier.mjs PLTR NVDA SOFI`) for one
     dossier each.
   - Takes ~5–15s per ticker. Auto-reads `AGENT_API_KEY` from `.env_agent_api`,
     hits production (TD Pro + TickerTrace), presents a browser UA.
   - Writes `runs/<TICKERS>_<date_time>/<SYM>_dossier.md`, `charts/<SYM>.png`, and
     `raw/<SYM>_*.json`. Prints the dossier to stdout and a final JSON line with
     `{outDir, results}`.

2. **Read the generated `<SYM>_dossier.md`** for the full tables.

3. **Open the chart PNG** with the Read tool and actually look at it — 90-day
   candles + EMA 8/21/55 + SMA200 + volume + RSI. Judge trend, structure, where
   price sits vs the fast EMAs (pullback-to-support vs extended vs broken). The
   visual read often **overrides the indicator table** — e.g. a name can look
   oversold on RSI but the chart shows a clean breakdown you shouldn't catch.

4. **WebSearch for recent news/catalysts** on the ticker — last few weeks: earnings
   (and the next earnings date), guidance, analyst rating/PT changes, sector news,
   anything that explains the current price. *News follows price* — find what moved
   it. (In any public/Substack write-up, the AI is "Sam," never "Claude.")

5. **Deliver the verdict to Michael — plain English, blunt, with a recommendation.**
   Don't dump the file. Answer the question he actually asked ("do you like it
   here?"). Structure:
   - **One-line take** — bull/bear/neutral and why, in a sentence.
   - **The setup** — where price is in its range, the trend, the chart read, the
     levels that matter (put wall / EMA21 / SMA200 as support; call wall / pivots /
     fib as resistance). Give a *where-would-you-buy* and *where's-it-wrong* level.
   - **What the options say** — flow lean (net premium direction), dealer
     positioning (max pain, walls, IV skew), expected move. Is smart money leaning?
   - **Ownership** — are funds adding or trimming? Flag if the "ownership" is mostly
     option-overlay/income ETFs (not real conviction).
   - **The catalyst / risk** — the news read; next earnings date; what could break
     the thesis.
   - **Bottom line** — would you buy it here, wait for a level, or pass. Be willing
     to say "no, not here." Push back; don't just validate the question.

## Notes / gotchas

- **This is a read, not a recommendation engine.** It aggregates data; the judgment
  is yours. Always reconcile the indicator table against the chart and the news —
  if they disagree, say so.
- **TickerTrace coverage is funds/ETFs**, including thematic and option-overlay
  products. High `fundCount` ≠ high conviction. Read providers: ARK/whale long-only
  adds mean something different than a YieldMax covered-call overlay.
- **No news endpoint** — the WebSearch in step 4 is not optional; without it the
  dossier is half-blind. Where price sits only makes sense next to *why*.
- **Flow timeframe** defaults to `week`. Override with `DD_FLOW_TIMEFRAME=today`
  (or `month`) for a tighter/wider flow window.
- **Auth/UA:** same plumbing as `stock-recap` — `AGENT_API_KEY` from
  `.env_agent_api`, browser UA (the edge WAF 403s the bare Node UA). The
  `ticker=` query param is what filters flow by symbol (`symbol=` is ignored).
- **Charts** reuse `stock-recap/.venv` + `render_chart.py`. If `stock-recap` isn't
  present or its venv is missing, the chart step is skipped and the data dossier
  still completes. Both skills live side-by-side in the repo's `.claude/skills/`.
- Each run is a fresh point-in-time snapshot; run folders are gitignored under
  `.claude/` and safe to delete.

From mphinance/alpha-skills, MIT.

More from mphinance/alpha-skills