Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Use when the user asks for an end-to-end orientation on a listed company they don't yet understand, especially when the request combines two or more dimensions in one ask. Triggers on "X 是干什么的", "帮我了解 X", "X 主营 + 同行", "盘前为什么涨", "本周趋势 + 阻力支撑", "X 和其他公司的关系", "first time looking at X", "full brief on X". Skip when the user wants only a single lens — use the targeted Longbridge sub-skill instead.
.claude/skills/kansoku-trade-stock-deep-dive/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-05 | ✗→✓ | ▲ Improved | 214% | 0% |
| case-13 | ✗→✓ | ▲ Improved | 57% | 0% |
| case-08 | ✓→✗ | ▼ Worse | 59% | 0% |
| case-09 | ✓→✓ | = Same ✓ | 149% | 0% |
| case-20 | ✓→✓ | = Same ✓ | 110% | 0% |
A six-lens onboarding workflow for getting up to speed on one listed company in a single pass. Built from a live session where the author repeatedly failed by quoting community-post numbers as if they were company guidance, by trusting GAAP EPS fields when the market quoted non-GAAP, and by reading YoY % deceleration as business slowdown without checking the base.
Core principle: anchor every numeric claim on a primary source. Press release, 8-K, real OHLCV are primary. Community topic titles, truncated news headlines, and aggregated provider fields are _leads_, not sources. Verify before quoting.
If the user wants only one lens (e.g., just real-time price), do NOT load this skill — go directly to the corresponding longbridge-* sub-skill.
Lenses 1–5 are independent. Dispatch their Longbridge calls in parallel. Lens 6 (audit) runs last, against all collected data.
| # | Lens | Sub-skills | Output anchor | | --- | -------------------- | ------------------------------------------------------------- | ----------------------------------------------------------------------------------------- | | 1 | Business identity | longbridge-company-profile + longbridge-business-query | 1-line "what" + segment revenue mix table | | 2 | Fundamentals | longbridge-fundamental | Quarterly revenue series (YoY _and_ QoQ) + OpInc trajectory + reconciled EPS | | 3 | Technicals | longbridge-technical + longbridge-kline | Last-5-day OHLCV + week summary + pivot S/R + indicator vote + ATR14 | | 4 | Catalysts | longbridge-news + trump-truth-monitor (if policy-exposed) | Classified news/filings + community sentiment skew + pre-market range + Trump-policy hits | | 5 | Supply chain & peers | longbridge-supply-chain + longbridge-competitive-analysis | Upstream → company → downstream flow + peer valuation table + paired-trade logic | | 6 | Narrative audit | (this skill, see below) | Reconciliation: official vs community, GAAP vs non-GAAP, YoY vs QoQ, mix vs aggregate |
These six errors occur repeatedly when synthesizing a stock brief. The skill exists primarily to prevent them. Self-audit before sending output:
Longbridge news <SYMBOL> returns mixed feeds. Items with URL pattern longbridge.com/topics/* are user-posted community threads, not company-issued. Any "FY guide $X" figure sourced from a community-topic title must be verified against the actual press release (longbridge.com/news/* with reputable source_name, or the SEC 8-K) before quoting. Never restate a community-topic number as guidance.
News titles end in "…" when truncated by the feed. Pull the article body (<url>.md variant on Longbridge) or the 8-K text before quoting a CEO percentage. If you can't pull the body, attribute as "per news headline — unverified".
financial-report --kind IS EPS field = GAAPUS companies report non-GAAP EPS on earnings calls; analyst consensus is also non-GAAP. The Longbridge IS EPS field is GAAP diluted. If the value looks absurd (e.g., 0.04 when consensus was 0.80), it's GAAP, not a beat/miss. Always state which basis you're quoting. For non-GAAP, pull from news / press release excerpt.
Compute QoQ alongside YoY every time. If the base year was itself accelerating off a low, YoY % naturally compresses even as absolute revenue accelerates. Look at the sequential QoQ trend before claiming "growth is slowing". Show implied forward YoY from next-quarter guidance — it often reveals a V-shape that aggregate YoY hides.
If the company has one hyper-growth segment and one declining segment (classic: AI silicon + legacy storage), aggregate growth understates the hot segment. Break out by segment when the disclosure allows; if not, estimate from CEO commentary and flag the assumption.
A pre-market range > 1× ATR14 (e.g., $188 – $211 vs ATR14 $12) means institutions and retail are in heavy disagreement. Note explicitly. The open print and the first-hour hold-of-pivot is the resolution.
# {Symbol} — {company name} 略览
## 一、Business identity
1-sentence "what" + revenue mix table (segment | % | content | cycle phase)
## 二、Fundamentals
- Quarterly revenue table: period | rev | YoY | QoQ
- OpInc trajectory (sign and direction matter more than absolute)
- EPS: GAAP from provider AND non-GAAP from press release (label both)
- Latest-quarter guidance vs prior guidance — explicit delta
## 三、Technicals
- Last 5 sessions OHLCV table + week summary (open/close/high/low/%)
- 52w range, last close, ATR14
- Pivot S/R from last 5d (P, R1-3, S1-3)
- Indicator vote table (MACD, RSI, KDJ, BB, EMA50/200, ADX, OBV)
- Composite verdict (buy / sell / neutral)
## 四、Catalysts (now)
- Classified news (catalyst / regulatory / strategic / opinion / filing / community)
- Pre-market last + range (flag if > 1× ATR)
- Sentiment skew (coarse %, no individual quotes)
## 五、Supply chain & peers
- Upstream (foundry / IP / EDA / equipment) — who, why dependent
- Downstream (customers / OEMs / end users) — concentration risk
- Horizontal (same-tier competitors / substitutes) — peer valuation table
- Paired-trade logic — which other tickers move together / inversely
## 六、Verdict
- Bull thesis (anchored on numbers)
- Bear / risk (anchored on numbers)
- Valuation anchor (which peer's multiple this should trade against, why)
- Key tell to watch (e.g., does first-hour price hold pivot)
⚠️ For reference only. Not investment advice.| Excuse | Reality | | ------------------------------------------------------- | ------------------------------------------------------------------------------------------- | | "The community post said the FY28 guide is $X" | Community ≠ company. Find the press release. | | "The CEO said +40% YoY" (from a "…"-truncated headline) | Pull the article body. Attribute as unverified if you can't. | | "EPS is 0.04, missed badly" (from the IS feed) | That's GAAP. Non-GAAP is what consensus measures. | | "YoY dropped from 63% to 22%, growth is slowing" | Check QoQ. Check the base. Check next quarter's implied YoY. | | "Revenue is +28%, no big deal" | If 75% of revenue is hyper-growth segment masking 25% in decline, the hot segment is +60%+. | | "Pre-market is volatile, ignore" | A pre-market range > 1× ATR is institutional disagreement — a signal. Note it. |
topics/* URLLenses 1–5 each call multiple Longbridge endpoints. Within a single user turn, batch every independent CLI call in one parallel tool block. A typical first turn dispatches 8–12 longbridge calls in parallel, then runs the technical Python computation in a follow-up turn.
Required (load on demand):
longbridge-company-profile, longbridge-business-querylongbridge-fundamentallongbridge-technical, longbridge-klinelongbridge-newslongbridge-supply-chain, longbridge-competitive-analysisOptional (deeper drilldown):
longbridge-peer-comparison — pure peer-matrixlongbridge-valuation — historical PE/PB percentilelongbridge-earnings — earnings-day specificlongbridge-capital-flow — intraday capital directiontrump-truth-monitor — Trump policy catalyst (run for lens 4 when the symbol has policy exposure: semis / China ADR / auto / energy / defense / bank)For session-long multi-symbol tracking after the deep dive, route to market-session-tracker.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 26,328 | 21,974 | -17% | 1 | 1 | 0% | 4,127 | 3,029 | -27% | 0 | 0 | — |
case-02 | fail→fail | 32,050 | 9,118 | -72% | 1 | 1 | 0% | 5,334 | 2,870 | -46% | 0 | 0 | — |
case-03 | fail→fail | 26,804 | 37,643 | +40% | 1 | 1 | 0% | 4,280 | 2,685 | -37% | 0 | 0 | — |
case-04 | fail→fail | 17,229 | 41,322 | +140% | 1 | 1 | 0% | 2,752 | 8,751 | +218% | 0 | 0 | — |
case-05 | fail→pass | 9,357 | 11,456 | +22% | 1 | 1 | 0% | 1,309 | 4,110 | +214% | 0 | 0 | — |
case-06 | fail→fail | 13,428 | 33,972 | +153% | 1 | 1 | 0% | 2,007 | 8,346 | +316% | 0 | 0 | — |
case-07 | fail→fail | 17,374 | 26,792 | +54% | 1 | 1 | 0% | 2,535 | 6,052 | +139% | 0 | 0 | — |
case-08 | pass→fail | 13,266 | 10,877 | -18% | 1 | 1 | 0% | 1,918 | 3,041 | +59% | 0 | 0 | — |
case-09 | pass→pass | 11,251 | 13,568 | +21% | 1 | 1 | 0% | 1,754 | 4,365 | +149% | 0 | 0 | — |
case-10 | fail→fail | 23,833 | 10,544 | -56% | 1 | 1 | 0% | 3,832 | 3,340 | -13% | 0 | 0 | — |
case-11 | fail→fail | 30,346 | 11,062 | -64% | 1 | 1 | 0% | 4,830 | 2,848 | -41% | 0 | 0 | — |
case-12 | fail→fail | 18,544 | 7,315 | -61% | 1 | 1 | 0% | 2,554 | 2,838 | +11% | 0 | 0 | — |
case-13 | fail→pass | 34,502 | 38,149 | +11% | 1 | 1 | 0% | 5,309 | 8,337 | +57% | 0 | 0 | — |
case-14 | fail→fail | 23,837 | 11,020 | -54% | 1 | 1 | 0% | 3,370 | 3,000 | -11% | 0 | 0 | — |
case-15 | fail→fail | 24,973 | 10,483 | -58% | 1 | 1 | 0% | 4,049 | 2,801 | -31% | 0 | 0 | — |
case-16 | fail→fail | 21,415 | 9,720 | -55% | 1 | 1 | 0% | 3,211 | 3,117 | -3% | 0 | 0 | — |
case-17 | fail→fail | 22,109 | 42,341 | +92% | 1 | 1 | 0% | 3,192 | 8,386 | +163% | 0 | 0 | — |
case-18 | fail→fail | 22,295 | 8,133 | -64% | 1 | 1 | 0% | 3,127 | 2,683 | -14% | 0 | 0 | — |
case-19 | fail→fail | 18,902 | 9,200 | -51% | 1 | 1 | 0% | 2,744 | 2,754 | +0% | 0 | 0 | — |
case-20 | pass→pass | 11,899 | 13,174 | +11% | 1 | 1 | 0% | 2,085 | 4,384 | +110% | 0 | 0 | — |
case-21 | fail→fail | 2,518 | 5,695 | +126% | 1 | 1 | 0% | 338 | 2,583 | +664% | 0 | 0 | — |
case-22 | pass→pass | 17,488 | 8,866 | -49% | 1 | 1 | 0% | 2,714 | 3,496 | +29% | 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 8 counted toward the lift figure. The other 14 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 +5 percentage points is the difference between those two pass rates over the 8 comparable cases. 5 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.