Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Use when reading today's US-market capital flow across multiple sectors to identify rotation direction — e.g. "今天资金流向", "板块强弱", "rotation map", "卖芯买云", "where is money moving today", "scan flows across sectors". Produces a cross-section snapshot of net inflows by cohort (indices / semis / software-cloud / mega-tech / AI applications), names the dominant narrative, and writes a dated journal file. Different from `market-session-tracker` (intraday live monitoring of a single watchlist) — this is a
.claude/skills/kansoku-trade-capital-rotation/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-07 | ✗→✓ | ▲ Improved | 82% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 25% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 121% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 111% | 0% |
| case-10 | ✗→✓ | ▲ Improved | 149% | 0% |
Scans capital flow across standard US cohorts in one session, identifies rotation direction, classifies winners / losers, names the dominant narrative, and logs a journal file.
> Scope: US-only. Do NOT query HK / CN / SG markets (user preference). > Sources: Longbridge capital, market-temp. Cite as 长桥证券. > Units: ambiguous — see TD-UNIT-01 in trading-discipline. Longbridge does not label the unit. Record the raw API number and the unit you inferred; do NOT silently convert (no 亿).
longbridge-capital-flow directly)market-session-tracker)| Cohort | Symbols | | -------------------- | ------------------------------------------------------------------------------------------------- | | Indices | SPY, QQQ, DIA, IWM | | Semis | NVDA, AMD, MU, MRVL, TSM, AVGO, SMH, SOXX, AMKR, ASX | | Software / Cloud | NOW, ORCL, CRM, ADBE, SNOW, DDOG, MDB, PLTR, PANW, CRWD, NET, IGV, CLOU | | Mega-tech | AAPL, MSFT, GOOGL, AMZN, META, TSLA | | Risk-off proxy | VXX, TLT, GLD (optional, for cross-asset confirmation) |
User watchlist override: read stocks/ directory for symbols the user already tracks; promote those to first-tier in their respective cohort.
date + confirm US session state (pre / intraday / post / closed). Adjust analysis date in filename: use the US session date, not Asia local date.bash longbridge market-temp US --format json
Report Temperature / Valuation / Sentiment.
bash longbridge capital SPY.US --format json longbridge capital QQQ.US --format json
Net large = capital_in.large - capital_out.large. Flag distribution if large net ≪ 0 while small net > 0 (主力—散户背离).
longbridge capital <SYM> --flow --format json | tail -8 to grab the latest cumulative inflow value (the last array element is the running total in 万 USD). Parallelize across symbols.~/git/trade/journal/YYYY-MM-DD-flow.md using the US session date. Use templates/rotation-snapshot.md as scaffold. If the file exists (e.g. re-run same day), append a new section with timestamp; do not overwrite.Use these triggers to label index behavior:
| Pattern | Label | | ----------------------------------------------------- | ----------------- | | SPY large net < 0 AND \|large net\| > 5 × small net | 机构派发 | | All 3 buckets (large / medium / small) net < 0 | 全档抛压 | | Large net < 0, small net > 0, magnitudes similar | 主力—散户背离 | | Large net > 0, small net < 0 | 主力吸筹 | | All 3 buckets > 0 | 全档吸金 |
Always state the pattern explicitly; do not say "weak / strong" vaguely.
A common useful narrative axis. Classify cohort flow winners / losers by AI revenue maturity:
When flow winners cluster in "已变现" and losers in "未变现", call out "narrative 收敛至 AI 已变现窄口" — this is a key macro signal of late-cycle AI selectivity.
bashlongbridge market-temp US --format json longbridge capital SPY.US --format json # snapshot (large/med/small) longbridge capital QQQ.US --flow --format json | tail -8 # time-series cumulative longbridge capital --flow --format json < SYM > .US | tail -8 # per-symbol
The --flow last-row inflow field is the cumulative net for the session in 万 USD. No date parameter — today's data only.
Error: request timeout / connect timeout → retry 1-2 times; do not block the report. Mark unavailable symbols with n/a and proceed..SOX.US) → substitute ETF proxy (SMH/SOXX).Tone: 中文白话, no jargon — see TD-LANG-01 / TD-LANG-02 in trading-discipline.
market-session-tracker — live intraday monitoring of one watchlistlongbridge-capital-flow — single-symbol drill-downlongbridge-market-temp — sentiment-only snapshotstock-deep-dive — multi-lens single-name researchcapital-rotation/
├── SKILL.md
└── templates/
└── rotation-snapshot.md| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 26,277 | 14,413 | -45% | 1 | 1 | 0% | 3,929 | 2,601 | -34% | 0 | 0 | — |
case-07 | fail→pass | 9,974 | 5,996 | -40% | 1 | 1 | 0% | 1,649 | 3,001 | +82% | 0 | 0 | — |
case-02 | fail→fail | 24,177 | 10,002 | -59% | 1 | 1 | 0% | 3,179 | 2,945 | -7% | 0 | 0 | — |
case-03 | fail→fail | 19,635 | 10,052 | -49% | 1 | 1 | 0% | 3,167 | 2,802 | -12% | 0 | 0 | — |
case-04 | pass→fail | 20,590 | 13,554 | -34% | 1 | 1 | 0% | 2,834 | 2,882 | +2% | 0 | 0 | — |
case-05 | fail→pass | 16,404 | 6,962 | -58% | 1 | 1 | 0% | 2,440 | 3,061 | +25% | 0 | 0 | — |
case-06 | pass→pass | 8,342 | 4,739 | -43% | 1 | 1 | 0% | 1,274 | 2,642 | +107% | 0 | 0 | — |
case-08 | fail→pass | 6,492 | 2,770 | -57% | 1 | 1 | 0% | 1,061 | 2,346 | +121% | 0 | 0 | — |
case-09 | fail→pass | 7,085 | 2,144 | -70% | 1 | 1 | 0% | 1,103 | 2,332 | +111% | 0 | 0 | — |
case-10 | fail→pass | 6,440 | 2,725 | -58% | 1 | 1 | 0% | 991 | 2,469 | +149% | 0 | 0 | — |
case-11 | pass→pass | 13,015 | 3,522 | -73% | 1 | 1 | 0% | 2,103 | 2,505 | +19% | 0 | 0 | — |
case-12 | fail→pass | 13,833 | 8,128 | -41% | 1 | 1 | 0% | 2,035 | 3,202 | +57% | 0 | 0 | — |
case-13 | fail→pass | 9,393 | 2,229 | -76% | 1 | 1 | 0% | 1,536 | 2,366 | +54% | 0 | 0 | — |
case-14 | pass→pass | 11,054 | 4,981 | -55% | 1 | 1 | 0% | 1,758 | 2,721 | +55% | 0 | 0 | — |
case-15 | pass→pass | 18,501 | 4,971 | -73% | 1 | 1 | 0% | 1,804 | 2,775 | +54% | 0 | 0 | — |
case-16 | pass→pass | 13,440 | 10,296 | -23% | 1 | 1 | 0% | 2,025 | 3,473 | +72% | 0 | 0 | — |
case-17 | fail→pass | 10,488 | 4,432 | -58% | 1 | 1 | 0% | 1,566 | 2,618 | +67% | 0 | 0 | — |
case-18 | pass→pass | 10,965 | 3,917 | -64% | 1 | 1 | 0% | 1,560 | 2,585 | +66% | 0 | 0 | — |
case-19 | pass→pass | 7,529 | 2,658 | -65% | 1 | 1 | 0% | 1,120 | 2,463 | +120% | 0 | 0 | — |
case-20 | pass→pass | 15,764 | 10,224 | -35% | 1 | 1 | 0% | 2,400 | 3,500 | +46% | 0 | 0 | — |
case-21 | fail→pass | 7,542 | 2,433 | -68% | 1 | 1 | 0% | 1,094 | 2,426 | +122% | 0 | 0 | — |
case-22 | fail→pass | 6,768 | 4,023 | -41% | 1 | 1 | 0% | 1,126 | 2,615 | +132% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 18 counted toward the lift figure. The other 4 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +41 percentage points is the difference between those two pass rates over the 18 comparable cases. 2 cases got worse with the skill loaded, and they are included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.