Market Session Tracker
Real-time US-market analysis pattern. Sits on top of longbridge-quote, longbridge-kline, longbridge-capital-flow, longbridge-market-temp — adds orchestration, breakout verification, distribution detection, tier classification, and revision discipline.
Standard symbol sets
| Theme | Symbols |
|---|---|
| Semi / memory | MU.US, TSM.US, DRAM.US (Roundhill Memory ETF), SMH.US, SOXX.US |
| Indices | QQQ.US, SPY.US, DIA.US, IWM.US |
| Vol / risk-off | VXX.US, UVXY.US, TLT.US, GLD.US |
.SOX.US is unavailable on Longbridge — use SMH/SOXX ETF proxies.
Seven protocols
0. Trump-feed sweep (pre-cash) — before any pre-market read, run python3 .claude/skills/trump-truth-monitor/scripts/fetch.py --hours 14. Any high-tier post touching watchlist sectors (tariff_trade / semi_tech / energy / fed_macro / geopolitical) goes into the session report as a candidate explanation for any gap, before running quote-based exuberance math. Skip when the watchlist has no policy-exposed names. See trump-truth-monitor skill for tier grading.
1. Pre-market verification — compute pre vol % of prev day full vol, and pre high % over prev_close. Flag exuberance when pre vol > 5% of prev day and pre high > prev_close × 1.07.
2. Failed-breakout 6-signal stack — count how many fire in the cash session:
- Pre-market high NOT touched in first 30 min of cash
- New intraday high breaks → price falls back below the broken level within minutes
- Volume does NOT expand at the breakout
- Sector ETF (
SMH/SOXX) does NOT confirm by going green - Leader stock (MU for memory; TSM for foundry; NVDA for AI) does NOT make new high
- Capital flow: all 3 buckets net selling
≥ 4 signals fired = failed breakout / distribution. Name the tier (§5).
3. Capital flow triple-bucket — longbridge capital <SYM> --format json. Net = capital_in − capital_out for each of large / medium / small. All 3 net out = textbook distribution. Use --flow for accelerating-outflow detection.
4. Cross-asset sentiment matrix
| Pattern | Interpretation |
|---|---|
| DIA > SPY > QQQ + VXX down | Rotation (defensive), not panic |
| VXX up + GLD up + TLT up | True risk-off |
| Sector red + SPY flat + VXX down | Isolated distribution |
| HK/CN valuation ≥ 80 + sentiment ≤ 35 | Known-bubble (overvalued, retail knows) |
5. Pullback tier classification
| Tier | Triggers |
|---|---|
| 1 震荡 | Stock −2% from intraday high; closes green |
| 2 实质回调 | Stock −5% from peak; sector ETF turns red |
| 3 板块下跌 | Sector −3%+; broad indices flat-to-red |
| 4 风险传染 | SPY −1%+; VXX +5%+; defensives also fall |
Always name the tier explicitly — never vague "weakening".
6. Scenario probabilities — always 3 scenarios (Bull / Base / Bear) with explicit % (sum=100) and trigger conditions. Mark probabilities as subjective. Revise as data flows with timestamps: 09:30 初判 → 09:54 修正 → 09:56 再修正.
7. Thesis revision discipline — when user says "突破"/"冲高"/"回调":
- Re-pull live quote + intraday minute tail — do NOT auto-agree
- Check cash intraday high vs pre-market high vs prior intraday high
- Distinguish: true breakout (new high > pre high, holds 5+ min) vs partial (breaks prior intra high but not pre high) vs recovery (only bounces from intra low)
- If data contradicts user, disagree with evidence
Output format (each snapshot)
- Time (ET) — always
- Symbol table — last, change%, intra high/low, vs pre high, vol
- Key signal (one sentence)
- Tier (if pullback context)
- Next watch levels (explicit prices, not "around X")
- Source: 长桥证券 · Disclaimer: ⚠️ 仅供参考，不构成投资建议
Session report logging
After a session, write a structured log using templates/session-report.md in this skill. Default path: ~/git/trade/journal/YYYY-MM-DD-<theme>.md (a dedicated git repo; journal/ avoids the logs global-gitignore collision). Captures pre-market verdict, opening behavior, tier evolution, thesis revisions, capital flow, cross-asset sentiment, end-of-day outcome, and lessons.
Optional: position context
When user provides positions or asks via longbridge positions:
- Show symbol, qty, avg cost, current price, unrealized P&L, % of book
- Cross-reference which positions are exposed to the current move
- Do NOT recommend buy/sell — defer to user
Anti-patterns
- ❌ Auto-confirming user's directional read (re-pull data first)
- ❌ Calling a cash bounce a "breakout" without checking pre-market high
- ❌ Single-point price prediction (use 3 scenarios)
- ❌ Vague "weak/strong" — use tier classification
- ❌ Conflating sector weakness with systemic (check VXX/GLD/SPY)
- ❌ Calling trend in first 5 min (wait for 30-min K)
