Imported from dadachundan/financial_agent (.claude/skills/catalyst-calendar/SKILL.md). Install upstream with npx skills add dadachundan/financial_agent --skill catalyst-calendar. Copyright stays with the author.

Catalyst Calendar

A single skill covering three distinct "what's coming up" lenses. Pick the mode from the user's phrasing — the modes share the same data sources but produce very different deliverables.

Language default (all three modes): English only. The Chinese companion is produced only on the explicit opt-in triggers listed under Mode C ("Report language (mode C)") — those triggers apply to Modes A and B too. A Chinese-language trigger phrase ("什么大事今天") does NOT by itself flip the output language.

Mode dispatch

If a request straddles modes (e.g. "what's big this week including any M&A votes?"), default to B (week-ahead) and surface the M&A votes inline in the milestone column — do not switch to mode C unless the user specifically asks for the single-deal deep dive.

Horizon dominates phrasing (dispatch tiebreaker). If the deliverable covers ≥3 trading days, includes a week calendar table, or includes a multi-day lookback recap, it is Mode B and saves to reports/calendar/ — even if the user literally said "morning note". A true Mode A brief is a single-session read; the moment it grows a "This Week's Calendar" table it has become a week-ahead and belongs in reports/calendar/<YYYY-MM-DD>[_<universe-slug>].md, not reports/morning/.

Mode A — Day-of brief

For "what's big today" / "what's hitting tomorrow" / "morning note" — a tight, opinionated brief readable in 2 minutes.

What to cover

Macro / data releases (priority 1 — these move the index)

• US: NFP / employment situation, CPI, Core PCE, FOMC decision, FOMC minutes, GDP advance/second/third, ISM Manufacturing / Services, Retail Sales, jobless claims (Thursday weekly), Consumer Confidence, U-Mich sentiment, JOLTS, PPI, Housing Starts, Existing/New Home Sales, Durable Goods, Industrial Production.
• China: CPI/PPI, official + Caixin manufacturing/services PMI, trade balance, industrial production, retail sales, fixed-asset investment, GDP, total social financing, M2.
• Eurozone: ECB rate decision, HICP flash, GDP, IFO, ZEW, eurozone PMI.
• Japan: BOJ rate decision, Tankan, CPI, GDP.
• Global: OPEC meetings, IEA monthly oil report, EIA crude/products inventory (Wednesday).

For each release: time (ET), consensus (Bloomberg/Reuters), prior, our view (if held), what would surprise the market.

Earnings (priority 2 — single-name moves)

• Pre-market: list ticker + consensus rev/EPS + 1-line "what matters" (key metric or guidance line PMs care about).
• After-close: same.
• For mega-cap reporters (AAPL, MSFT, NVDA, GOOG, META, AMZN, TSLA), give a sentence on read-across to the index / sector.

Fed / central bank speakers (priority 2)

• Time, speaker, topic if known, whether voter on current FOMC. Skip non-voting regional Fed presidents giving routine remarks unless they're on a tape that traders watch.
• Tag the stance: hawk / watchful / dove + expected emphasis (inflation / labor / financial-conditions). See Broker-style methodology § B Fed speaker hawk/dove map.

M&A milestones (priority 3 — situational)

• Any deal-specific milestone hitting that day: shareholder vote, HSR expiry, EU phase-I deadline, expected close, regulator decision date.
• For each: deal name, milestone, expected market reaction (target spread tightens/widens; arbs unwind).

Other

• Index rebalances — carry the effective date + per-sector passive-flow impact, not a one-liner. Model on GS's EM Flows Monitor: "FTSE China A50 rebalance effective 6/18; estimated passive inflow to hardware-semis / consumer, outflow from insurance / energy." List each rebalance (S&P, Nasdaq-100, MSCI, FTSE) with its EFFECTIVE DATE and the estimated in/outflow by sector. Any flow-impact figure cites the index provider's rebalance announcement (deep URL) per citation / numerical-accuracy rules.
• Options expiry (monthly OPEX = 3rd Friday; quadruple witching), Treasury auctions if mega-size, government shutdown / debt-ceiling deadlines, geopolitical (election results, summits with potential market impact).

Market-pricing context (lighter than Mode B — pick 1–2 signals)

• For the day's main event: SPX 1DTE straddle implied move (as-of timestamp) + VIX entering AM + FedWatch probabilities if relevant. One line per signal.
• See the Event positioning lens section for the full toolkit. Day-of briefs use a tight subset; the Mode B preview is where the full per-indicator playbook lives.

Output format

Keep it scannable. A morning desk read is one screen, not three.

# Morning brief — <YYYY-MM-DD>

**Top thing:** <one-line headline — what dominates the tape today>

## Macro calendar (ET)
- 08:30 — <release> — consensus X, prior Y. <one-line why-it-matters>
- 10:00 — <release> — same.
- 14:00 — FOMC <minutes/decision> — <what's already priced>.

## Earnings
**Pre-market:** TICKER (cons rev $X / EPS $Y) — <one-line focus>; TICKER2 (...) — <...>.
**After-close:** TICKER3 (...) — <...>.

## Fed / central bank speakers
- 09:30 — Powell, <topic>, <voter status>.
- 13:00 — Lagarde, ECB testimony.

## M&A / deal milestones
- ANSS shareholder vote (re: SNPS deal); current spread <X%>, expected to pass on offered premium.
- HSR expiry on <DEAL> at midnight — no second request expected based on no-issue letter <date>.

## Other
- Monthly OPEX (3rd Friday); $X notional in SPX, gamma <neutral/long/short>.
- Treasury 10Y auction 13:00 — recent auctions have tailed <N>bps.

## Trade angles (optional — skip if nothing actionable)
- Long/short <ticker>: thesis + catalyst + risk.

---
*Time stamp: <local time>. Pre-market may move by open.*

Day-of guardrails

• Be opinionated — a morning note that just lists events without a view is useless. Lead with the most important thing.
• "No news" is a valid brief — if nothing material is on the calendar, say so ("nothing material on the macro tape; Q3 earnings season effectively wound down; quiet day, expect chop on flows") and stop. Don't pad.
• Distinguish actionable from noise — major data release + mega-cap earnings + FOMC decision = actionable. A regional Fed president talking on community banking ≠ actionable for most desks.
• Time-stamp the take — if writing at 06:30 ET, note pre-market moves may invalidate it by 09:30.
• Don't fabricate consensus numbers. If you can't pull the Bloomberg/Reuters consensus, say "consensus pending" and link the calendar source (e.g. Investing.com economic calendar, BLS release schedule, BEA release schedule). Don't make up a number that sounds plausible.
• Macro release times in ET unless the user specifies otherwise. China releases in Beijing time + ET equivalent.

Day-of output location

Default: in-chat only. Save to a file only when the user says "save it" / "write this up" / "add to morning folder" / "保存".

When saving: reports/morning/<YYYY-MM-DD>_<topic-slug>.md (matches existing reports/morning/ directory convention — flat, no per-day sub-folder).

Pre-save checklist (saved Mode A briefs only): horizon really is ≤2 sessions (else it's Mode B → reports/calendar/, per the dispatch tiebreaker); no fabricated consensus / house decimals; all newly added URLs HTTP-checked (including xs-macbook-air.local); verification log appended per the cross-cutting guardrail. Re-run on any in-place update.

Mode B — Week-ahead horizon calendar

For "what's coming this week" / "catalyst calendar" / "upcoming events" — a multi-day table view over a coverage universe.

Step 1: Scope

Resolve from the request:

• Coverage universe — explicit ticker list, sector (e.g. "semis"), or user's tracked portfolio.
• Horizon — next 5 trading days (default), next 2 weeks, this month, this quarter.
• Include macro? — default yes; user can scope to ticker-specific events only.

Step 2: Gather catalysts

For each company / sector / macro lane:

Earnings & financial events

• Quarterly earnings date + time (pre / after market); consensus + buyside whisper if known.
• Annual shareholder meeting, investor day, capital markets day.
• Debt maturity / refinancing dates large enough to be a story.
• Earnings-preview brief structure per § M: consensus + sub-segment forecast + capacity/delivery proxies + contract-structure changes + delivery-timing trends + platform partnerships + read-across implications. Populate ≥4 signal types per name beyond the rev/EPS line.
• Alt-data leading indicators per § J — when an alt-data series breaks a multi-quarter trend, surface it with the segment-level forecast (AWS web traffic → cloud, OpenTable → restaurants, TSMC monthly → foundry customers, etc.).

Corporate events

• Product launches, FDA AdComm / PDUFA dates, regulatory decisions, contract expirations, lockup expirations, management transitions, dividend ex-dates (for income-portfolio context).

M&A milestones (the calendar version of mode C — surface the date, not the deep analysis)

• Shareholder votes, HSR expiry, EU phase-I/II deadlines, SAMR review windows, expected close dates.
• Optional market-level M&A-cycle read (distinct from the single-deal milestone above) — fold GS "Deal Momentum Builds" as a Mode B macro input: the cycle call ("global M&A +18% next 12m") and an acquisition-probability target basket (">=15% probability names"). This is a market-cycle input, not a single-deal monitor — keep it separate from Mode C, which has no direct sell-side merger-arb-monitor analog.

Industry events

• Major conferences with company schedules (CES, JPM Healthcare, Money 20/20, MWC, Computex, GTC, Hot Chips, ISSCC, RSA, Dreamforce, re:Invent, etc.).
• Trade shows, industry data releases (monthly auto sales, weekly box office, etc.).
• Index rebalances with effective date + per-sector passive-flow impact — list each (S&P, Nasdaq-100, MSCI, FTSE) with its EFFECTIVE DATE and estimated per-sector in/outflow, GS EM-Flows-Monitor style (see the Mode A "Other" rebalance pattern). A recurring quantified catalyst, not a one-liner.

Macro events

• US: FOMC, NFP, CPI, PCE, ISM, retail sales, jobless claims — week-ahead view.
• Non-US: ECB, BOJ, BOE, PBOC, SAMR; CPI, PMI, GDP releases for major economies.
• Geopolitical: G7/G20, elections, summit deadlines.
• For each macro print, pull the sub-component watch list from Broker-style methodology § A — track core MoM / supercore / shelter for CPI; trade services / final-demand-energy for PPI; wage growth / revisions / 3M MA for NFP; etc. Map each sub-print to a causal mechanism, not just "hot vs cold."
• Apply threshold-based historical rules from § K — if a level crosses a rule's threshold (CPI > 4% YoY, NFP > 125k 3M MA, etc.), flag the historical pattern explicitly in the preview.
• Score qualitative Fed publications when in the window (Beige Book, minutes, Powell pressers) using the § L LLM-as-tool framework — track Δ vs prior release on inflation / labor / growth / recession-concern / FCI dimensions.
• Tariff / regulatory schedule — list discrete dated events ticking inside the window (Section 301 expiries, FDA AdComm dates, EU Phase II deadlines, M&A vote dates). See § G.

Market-pricing context (priority 2 — pair with every H-impact event)

For each high-impact macro print or earnings name, capture the option / VIX / spread context so the bull/bear thresholds are actionable, not abstract. Use the Event positioning lens section below as the toolkit reference. Concretely, per H-impact event:

• Macro: SPX 1DTE straddle implied move (as-of), VIX + VIX1D entering the week, FedWatch path probabilities, 2y yield level, MOVE percentile if rates vol is the relevant axis.
• Earnings: ATM straddle implied move (Friday-expiry post-print), 3-yr historical-realized vs implied (long-gamma-cheap or rich?), single-name IV vs sector mean.
• Cross-asset cross-check: HY OAS + 5y5y breakeven — do credit and inflation expectations agree with the equity-vol read?

If the data is not pullable at write time (e.g. cron run on a closed market), say "implied move pending market open" rather than fabricating.

Completeness cross-check (mandatory before rendering Step 3). A week-ahead that misses scheduled releases is worse than no calendar — past failure: a Monday was labelled "(quiet)" while the same-week Nomura weekly listed the NY Fed Survey of Consumer Expectations (directly on the report's own theme) plus five other releases, and the 3-yr Treasury auction was dropped even though the 10y/30y legs made the table. Before rendering the table:

1. Pull a full-week release schedule — BLS release schedule + BEA release schedule + a consolidated economic calendar (e.g. Briefing.com) — and enumerate every scheduled release as a table row. Second-tier prints (NY Fed SCE, NFIB, trade balance, wholesale inventories, mortgage apps, budget statement, all Treasury auction legs) get one-line L-impact rows, not omission.
2. When present in zsxq, cross-check against the latest Nomura / GS week-ahead forecast table (the Nomura weekly's "Forecasts for economic indicators released during the week of …" figure) and add anything it lists that the sweep missed.
3. A day may be labelled "(quiet)" only after this cross-check, and its Notes cell must say what was checked: no releases per BLS/BEA/Briefing.com sweep YYYY-MM-DD.

Step 3: Calendar view

Render as a sortable table:

| Date | Day | Time (ET) | Event | Company/Sector | Type | Impact | Expectation | Notes |
|------|-----|-----------|-------|----------------|------|--------|-------------|-------|
| 2026-06-09 | Mon | 10:00 | ISM Services | Macro | Macro | M | downside-risk | Cons 53.0; house per Nomura 51.5 [PDF link] — new-orders sub-index rolling; <50 = recession-concern flare |
| 2026-06-10 | Tue | AMC | Q2 earnings | ORCL | Earn | H | meaningful beat possible | Cons rev $19.1B; DB models OCI ~$5.5B (+89% YoY) vs cons 92% [PDF link] — capacity-constrained = bullish demand |
| 2026-06-11 | Wed | 14:00 | FOMC decision | Macro | Macro | H | binary | FedWatch 25bp cut ~70% priced; dots in focus |
| 2026-06-11 | Wed | 14:30 | Powell presser | Macro | Macro | H | in-line | Tone on cut path |
| 2026-06-12 | Thu | 08:30 | CPI | Macro | Macro | H | upside-surprise likely | Cons 0.2% MoM core; house per Nomura 0.18% [PDF link] — shelter softening, IT-hardware tariff pass-through offsetting |
| 2026-06-13 | Fri | — | Triple witching | Index | Other | M | n/a | $X notional |

Three distinct dimensions — never conflate them (the brokers always carry them separately; Morgan Stanley's "Catalyst Preview: What's Ahead?" tags every catalyst with importance and expectation):

• Impact column — does the index move? H = market or major-sector move likely; M = single-name move likely; L = routine, useful to know.
• Importance (optional desk-vocabulary tag inside Notes when distinct from Impact) — does the thesis care? "very important" / "high" — a name can be low-Impact but very-important to a specific coverage thesis.
• Expectation column — which way do WE lean? One of {in-line / upside-surprise likely / downside-risk / meaningful beat possible / binary / n/a}. This is the analyst's surprise-direction view, NOT the index-impact rank.

Notes-cell convention for macro rows (high-priority broker pattern). Mirror GS "US Week Ahead" and Nomura "US Economic Weekly": carry a committed house forecast vs consensus PLUS the sub-component driver inline, not just "Cons X." Wire in the § A sub-component watch list: e.g. Cons 0.2% MoM core; house per Nomura 0.18% [PDF link] — shelter softening, IT-hardware tariff pass-through offsetting. The house number must be one of two honest things: (a) an adopted, attributed broker forecast — house per Nomura: 0.183% [PDF link] — taken from the freshest zsxq weekly, or (b) an explicit inline derivation whose inputs are each cited — our ~0.2% = shelter trend [link] + energy fade [link]. An unattributed bare our X.XX% decimal is forbidden — Claude has no forecasting model, so an invented decimal is fabrication. Cite the consensus source separately; never present any estimate as if a URL contained it. Any expectation tag quoting a number must cite a source containing that number, per CLAUDE.md § "Numerical Accuracy".

Step 4: Weekly preview note

Markdown narrative companion to the table:

# Week of <YYYY-MM-DD>: catalyst preview

**The week in one paragraph.** Headline view of what dominates: rates? earnings? a specific binary event?

## Key events ranked

1. **<Day> — <Event>** — why it matters; what's priced; how the print would surprise.
2. **<Day> — <Event>** — same.
3. ...

## Earnings of note

- **TICKER (Day, time)** — what we're watching: <metric>. Consensus rev/EPS. Recent stock action into print.
- ...

## Macro

For each H-impact release, structure the entry as:
- **Release / time / consensus / prior** — the basics.
- **House forecast vs consensus (adopted or derived — never invented)** — the GS / Nomura product is the *house number next to Street*, never hidden behind "consensus 0.2%." Get the decimals honestly: **adopt** the freshest zsxq broker forecast with attribution (`house per Nomura: 0.183% [PDF link] vs cons 0.20%`), or **derive** one inline with every input cited (`our ~0.2% = shelter trend [link] + energy fade [link]`). A bare unattributed `our 0.18%` is fabrication — forbidden. Cite the **consensus** source separately.
- **Sub-component forecast with directional bias** — broker-grade: e.g. "core CPI MoM 0.18% (vs 0.30% prior); shelter softening, IT hardware tariff pass-through offsetting." Reference the [sub-component watch list](#a-sub-component-watch-lists-for-major-macro-prints).
- **What would surprise** — the specific level + direction that moves the tape.
- **Cross-asset confirmation** — does rates vol / credit / FX / inflation expectations agree? [§ C cross-asset framework](#c-cross-asset-confirmation-framework).
- **Curve trade or central-bank meeting probability** if relevant — per [§ D](#d-curve-trades--central-bank-meeting-probabilities).

## "Forward data" calendar with directional bias

Day-by-day list per [§ E](#e-forward-data-calendar-with-directional-bias) — every entry has a directional bias + mechanism, not just a date and consensus.

## Fund flows + earnings revisions (standard weekly block)

GS runs Fund Flows as a *standalone weekly product* — this is a first-class block, not an afterthought. Dollarize and direct every figure (GS Weekly Fund Flows format), never "flows were positive":
- **Fund flows** — equity-vs-bond inflows by region ($), sector skew, cross-border FX flows, retail margin / leverage, hedge-fund net-positioning deltas. Standard read: "$23bn global equity inflows, US drove; KR −$12.1bn." State the source (EPFR-style / GS flow tables) inline; if flows aren't pullable at write time, say "flows pending" rather than fabricate.
- **Earnings revisions** — forward FY EPS revised YTD by sector; "revisions leading vs lagging the index" tells you cycle stage (revisions leading = healthy rally; index leading revisions = late-cycle).
- See [§ H](#h-fund-flows--earnings-revisions-trackers).

## Positioning implications

For each H-impact event:
- **What's priced** — FedWatch / 2y yield / consensus path. One line.
- **Implied move** — SPX 1DTE straddle or single-name straddle (with as-of timestamp).
- **Vol-surface tell** — VIX vs historical event-day average; VIX1D backwardation vs contango; SKEW; VVIX. Flag if cheap or rich relative to setup.
- **Cross-asset check** — HY OAS + MOVE + 5y5y breakeven agree or disagree with equity vol?
- **Decision** — long gamma / short gamma / directional / flat. One sentence with the trigger that would flip the call.

Apply the [Event positioning lens](#event-positioning-lens--option--vix--spread-context-modes-a--b) per-indicator playbook table — each indicator type has a different primary tool.

## Next week heads-up (forward-roll with our forecast vs consensus)

Every GS "US Week Ahead" and Nomura weekly closes with an explicit forward-roll that states the bank's OWN forecast vs consensus on the headline upcoming print — not just "CPI is next week." Mirror it:
- <One-line> on the biggest event already on the radar for the week after, with `house per <broker> X [PDF link] vs cons Y` (or an inline cited derivation) on the headline print — same adopted-or-derived rule as the Macro block; never a bare invented `our X`.

Dedupe rule (one trade, one home). Each trade decision gets ONE full statement, and the Positioning implications table is its single home. The TL;DR / key-takeaways block may carry a one-line version; every other section references it ("see Positioning implications") instead of restating the rationale. No section may re-derive a conclusion already derived elsewhere — past failure: the same sell-vol/IV-crush call restated ~10 times across one weekly. Benchmark broker weeklies (GS Kickstart, Nomura) are rigidly templated; each call appears once in its slot.

Step 4b — Market/region preview (GS Kickstart template)

When the universe is an index or region (e.g. "what's coming for the S&P / Taiwan / EM next week?") rather than a ticker list, the per-name calendar table doesn't fit — use the GS Weekly Kickstart skeleton instead. It is rigidly templated, which makes it diff-able week to week. The six blocks, in order:

1. Index move + sector winners / losers — week's % move, the OW/UW sector dispersion.
2. Fund flows $ by region + sector skew — dollarized per the § H format ($23bn global equity inflows, US drove; KR −$12.1bn).
3. Forward EPS-revision direction — by sector; revisions leading vs lagging the index.
4. Valuation percentile vs own ~10y history — "fwd P/E at the 88th percentile of its own 10y range," not an absolute multiple in a vacuum.
5. 3 / 6 / 12-month index target — with the EPS + valuation bridge (target = forward EPS × target multiple; show both inputs).
6. OW / MW / UW sector grid — the desk's allocation tilts.

Every percentile / flow / target number traces to a source per CLAUDE.md § "Numerical Accuracy." Any flows or valuation-percentile chart carries the in-chart source annotation per the global chart rule. This template complements (does not replace) the ticker-level Step 3 table — use whichever matches the universe.

Step 4c — Exhibits (default for file-saved weeklies)

Broker weeklies are exhibit-heavy (Nomura carries Figs 1–10 with in-chart "Source: BLS, Haver, Nomura" footers; GS Kickstart is ~24 pages of exhibits) — a file-saved Mode B weekly that quotes VIX/MOVE/HY-OAS levels in tables but renders zero charts under-delivers. Default: 2–4 matplotlib PNGs generated from db/indicators.db (read-only SELECTs only; /opt/anaconda3/bin/python3), e.g.:

• Vol-complex panel — VIX / VIX1D / vix_slope, ~3-month window;
• CPI core-MoM trend with the upcoming print date marked;
• Claims 4-wk MA vs its band;
• HY OAS vs VIX disagreement chart (the cross-asset dissenter visual).

Save to reports/charts/calendar_<week>_*.png and embed. Every chart obeys the global chart rules: in-chart source footer ("Source: CBOE/FRED via db/indicators.db, as of YYYY-MM-DD"), data covering the full visible span, rightmost point fresh, x-axis clipped to the intersection of plotted series. Skip exhibits only when the deliverable is an in-chat 1–2-day horizon.

Week-ahead output location

reports/calendar/<YYYY-MM-DD>.md where the date is the literal Monday-of-week date (e.g. 2026-06-08.md). An optional universe slug keeps a sector week-ahead and the macro week-ahead from colliding under the update-in-place rule: 2026-06-08_semis.md. Create the directory if missing. Update in place if a previous pass for the same week + universe exists.

Short horizons (1–2 days) can stay in-chat; only save when the horizon is ≥3 days or the user says "write this up".

Pre-save checklist (Mode B — run before EVERY save, including in-place updates)

Mandatory blocks accrete across skill commits and silently drop out of outputs — so the full list lives here, in one place. Tick every line before writing the file:

• Step 3 table carries the Expectation column (Impact alone is not the spec).
• Macro Notes cells carry house-vs-cons + sub-component driver (adopted/derived per the house-number rule — no bare our X.XX%).
• Completeness cross-check done (Step 2); any "(quiet)" day names the sweep that cleared it.
• Fund flows + earnings revisions block present (or explicit "flows pending").
• Forward-roll ("Next week heads-up") states house-vs-cons on the headline upcoming print.
• Every borrowed broker PT carries report-date price + implied upside per § M ("DB: Buy, PT $300" bare is a violation).
• If ≥2 zsxq notes were cited on the same name / print: every institute view dated, same-institute revisions shown old → new with trigger, disagreements flagged side-by-side per § M Sell-side view evolution (卖方观点演变) — never blended.
• Exhibits present per Step 4c (or the horizon is in-chat 1–2 days).
• Dedupe rule honored — each trade decision stated fully once, in Positioning implications.
• Further viewing block present, or an explicit "nothing to visualize this week" line.
• Verification log appended per the cross-cutting guardrail.
• All newly added URLs HTTP-checked — including xs-macbook-air.local links (see cross-cutting guardrails).
• Word count within budget (hard cap 5,000).

Any in-place update to an existing reports/calendar/ or reports/ma/ report — even a link fix — re-runs this checklist before commit. A touched report that still misses a mandatory block is a fresh violation, not grandfathered.

Mode C — Single-deal M&A monitor

For "track the X-Y merger" / "what's the spread on Z" / "M&A status on " — a deep, single-transaction report.

Deliverable: a 3,000–6,000 word English markdown report on one active or recently-closed M&A transaction. The report answers five specific questions:

1. What's the deal? Target, acquirer, consideration, implied value per target share, announcement date, expected close.
2. What's the spread? Current target price vs deal value; annualized return assuming the expected close date.
3. Where are we on the milestone path? Shareholder vote, antitrust approvals, financing condition, other closing conditions.
4. What's the break risk? Financing / antitrust / shareholder / litigation / acquirer-stock / macro.
5. What's the probability range and what would change it? Bear / base / bull paths with named triggers.

Adapted from the LLMQuant M&A event tracker (MIT), re-pointed at SEC EDGAR + web search for U.S. deals and cninfo / HKEX / TDnet for cross-border deals.

On precedent: the sell-side library this skill draws from has no direct analog for a single-deal merger-arb monitor — the bank M&A coverage there is single-name reaction notes and market-level M&A-cycle strategy (e.g. GS "Deal Momentum Builds"), not pending-deal spread / milestone / break-risk trackers. Mode C's EDGAR-anchored rigor stands on its own; do not imply broad sell-side precedent for the merger-arb format. (The market-level M&A-cycle read belongs in Mode B as a macro input, not here.)

When to use mode C

The user says any of:

• "Track the SNPS-ANSS merger"
• "What's the deal spread on TTWO-Codemasters?"
• "M&A status on AVGO-VMW closing"
• "Is the CDNS-Hexagon D&E deal closing?"
• "Risk of break on the X-Y deal?"
• "Track the antitrust on TSMC-Intel JV"
• "Merger arb on Activision-Microsoft" (post-mortem also fine)

Supports:

• Pending deals (announced but not closed) — primary use case; produces spread + probability.
• Pre-announcement speculation (rumored deals with credible sourcing) — produces a deal-likelihood map; explicitly labels as speculation; no spread math.
• Recently closed deals (last 6 months) — produces post-mortem analysis of how the deal went vs the consensus pre-close view.

When NOT to use mode C

• Routine quarterly earnings — use [[earnings-analysis]].
• Pure regulatory / antitrust monitoring with no specific deal attached — use [[regulatory-risk-monitor]].
• Private-to-private transactions with no public disclosure — there's nothing to anchor the analysis on; decline.
• Spinoffs / reverse-Morris-trust splits — these are M&A-adjacent but follow a different filing pattern (Form 10, when-issued trading); skill applies imperfectly. Flag the limitation.
• "What deals are happening this week?" — that's mode B (week-ahead), not mode C; the milestone column lists deals by date.

Core principle: accuracy over completeness — never hallucinate

The accuracy rules from [[company-research]] apply verbatim. M&A-specific failure modes:

• Never invent a deal term. Consideration mix (cash %, stock %, ratio), implied price, walk-away conditions, termination fee, expected close date — every one of these must come from a specific SEC filing (S-4, DEFM14A, 8-K Item 1.01, 425) or a verifiable press release URL.
• Never compute a spread without current target price. "Spread is 4.5%" requires: deal-implied value per share, current target price (with date), and the assumed days to close. Show all three.
• Never quote a closing date the company hasn't stated. "Expected to close in 2Q26" must trace to the deal proxy or a company press release. "Sometime in 2026" is fine if that's what the company said; "by April 2026" is not unless that's literally in a filing.
• Never call a deal "likely to close" or "unlikely to close" without addressing the four standard break risks (financing, antitrust, shareholder vote, litigation). A probability statement that ignores even one of these is incomplete.
• Never source antitrust language to a press release when the actual filing is public. DOJ / FTC consent decrees, EU Commission decisions, and CMA / SAMR filings all have permanent URLs — cite the docket, not a Bloomberg article.
• Never confuse "merger consideration" with "fair value." The deal price is what the buyer is paying; fair value is the target's standalone value. They are different and must not be conflated.

Report language (mode C)

Default behavior: English only. This is a monitoring / tracking skill, not a deep-research deliverable — most users want the English read and don't need the Chinese companion every time. (The substantive research skills company-research / compare-companies / earnings-analysis / sector-overview still default bilingual; this mode does not.)

Chinese opt-in (any of these triggers a Chinese companion file alongside the English):

• also in Chinese / add Chinese / bilingual / both languages / --bilingual / --zh
• 用中文也输出一份 / 也输出中文版 / 中英双语

Chinese-only (skip English): 用中文即可 / --zh-only / Chinese only.

When a Chinese companion is produced, use bilingual technical terms: M&A / 并购, target / 标的, acquirer / 收购方, consideration / 对价, implied value / 隐含估值, spread / 套利价差, expected close / 预计交割, termination fee / 终止费, antitrust / 反垄断, shareholder vote / 股东投票, definitive proxy / 最终代理征集书. Keep ticker codes, regulator names (FTC, DOJ, SAMR, MOFCOM, CMA, EU), and case numbers in original form.

Filenames (mode C):

• English: reports/ma/<Target>_<Acquirer>_<YYYY-MM-DD>.md (e.g. ANSS_SNPS_2026-05-31.md)
• Chinese: reports/ma/<Target>_<Acquirer>_<YYYY-MM-DD>_zh.md
• Use English/pinyin name (per the project filename rule), not Chinese-only. Mixed-domicile deals: <English-Target>_<English-Acquirer>_<YYYY-MM-DD>.md (e.g. Hexagon-DE_Cadence_2026-05-31.md).

Update-in-place rule: one report per ordered (target, acquirer) tuple. If a report from a prior tracking pass exists, update it in place (refresh the date suffix; git history records the trail). Do not pile up dated copies.

Data sources (mode C)

Primary (SEC EDGAR — US deals)

For every U.S.-domiciled deal, pull the following from EDGAR. Resolve via the submissions JSON (https://data.sec.gov/submissions/CIK<padded>.json):

For the user's project: helper at fetch_financial_report.py (DB db/financial_reports.db) covers 10-K/10-Q/8-K. M&A-specific extensions for S-4 / DEFM14A / 425 may need a one-off pull — use the same EDGAR submissions JSON pattern. Save extracted M&A docs under oneoff/ma_<TARGET>_<ACQUIRER>/.

Primary (non-US deals)

• Chinese A-share / HK targets → cninfo (巨潮资讯) + HKEX news room. Look for: 要约收购 (tender offer), 资产重组报告书 (M&A restructuring report), 重大资产重组 (material asset reorganization), 股东大会通知 (shareholder meeting notice). Helper: fetch_cninfo_report.py.
• Japanese targets → EDINET (Yuho updates), TDnet (decision-day press release in 「公開買付届出書」 / tender offer registration statement), MOJ for antitrust filings.
• Korean targets → DART (주요사항보고서 material disclosures, 타법인주식 및 출자증권 양수결정 acquisition decisions).
• UK targets → London Stock Exchange RNS + Takeover Panel announcements; UK Code on Takeovers and Mergers governs disclosure cadence.
• EU cross-border → European Commission Merger Regulation (M.xxxx case numbers), national antitrust authority filings (Bundeskartellamt for Germany, AGCM for Italy, etc.).

Antitrust / regulatory tracking

• U.S. DOJ Antitrust Division → press releases at justice.gov/atr and CMS for case filings.
• U.S. FTC → press releases at ftc.gov/news-events/press-releases and Hart-Scott-Rodino filings (premerger notification — the company files but the receipt is not public; FTC announces second requests if issued).
• European Commission DG-COMP → case search at ec.europa.eu/competition/elojade/isef/index.cfm with the M.xxxx case number.
• UK CMA → gov.uk/cma-cases with case ID.
• China SAMR (State Administration for Market Regulation) → press release archive; concentration cases categorized as 经营者集中.
• China MOFCOM → for legacy / foreign-investment-overlap cases.

Pricing and spread

• Target current price + history — yfinance auto_adjust=True.
• Acquirer current price + history — same.
• Deal-implied value per target share — computed from announcement terms (cash + stock ratio × acquirer current price).
• Spread = (deal-implied value − target current price) / target current price.
• Annualized return = spread × (365 / days to expected close).

Workflow (mode C)

Step 0 — Parse inputs and CONFIRM target/acquirer direction

Critical: never silently assume which company is the target and which is the acquirer. "SNPS-ANSS" reads either direction; getting it wrong corrupts the entire deal-snapshot block (the spread math swaps, milestones swap, break-risk attribution flips). The skill must resolve direction explicitly before writing anything.

Three acceptable input forms (in preference order):

Direction-resolution workflow:

1. If the user's phrasing is explicit ("X is acquiring Y" / "X acquired Y" / "target=A acquirer=B"), proceed directly.

2. If only one ticker is given ("track the ANSS deal", "M&A status on AVGO"), use EDGAR to auto-detect:

   • Pull the named ticker's last 12 months of 8-K Item 1.01 ("Entry into a Material Definitive Agreement") and Item 2.01 ("Completion of Acquisition or Disposition").
   • The 8-K language explicitly states "Agreement and Plan of Merger" with the counterparty named, and identifies which side is being acquired.
   • Resolve the counterparty's ticker via EDGAR's ticker→CIK map.
   • Surface the detected direction to the user before writing ("Resolved: ANSS = target, SNPS = acquirer, per ANSS 8-K Item 1.01 filed 2024-01-16. Confirm before I proceed?").

3. If two tickers are given ambiguously ("SNPS-ANSS", "AVGO-VMW"), do NOT assume order — ask the user:

   • "Which side is the target (being acquired) and which is the acquirer? E.g. for SNPS-ANSS: did SNPS acquire ANSS, or did ANSS acquire SNPS?"
   • Once the user confirms, proceed with the verified direction.

4. Optional safety check even after explicit confirmation: pull the target's 8-K Item 1.01 and verify the counterparty named in the filing matches the acquirer the user stated. If they disagree, surface the conflict to the user before writing — the filing is authoritative.

Resolve both sides to:

• Ticker + exchange (or Private if unlisted).
• CIK (via EDGAR ticker→CIK map at https://www.sec.gov/files/company_tickers.json).
• Domicile (drives which portal to use as primary source).

Filename convention is target-first (reports/ma/<Target>_<Acquirer>_<YYYY-MM-DD>.md) — but this is now a consequence of the resolved direction, not a request-format assumption. The user does not need to put target first in the request; the skill does the resolving and writes the filename correctly.

If the deal hasn't been announced yet ("rumored Microsoft-Sony deal"), explicitly label the report as "Speculative — no definitive agreement disclosed" and skip the spread math. Direction confirmation still applies — even a rumored deal has a presumed acquirer and target.

Failure modes:

• Silently assuming "first ticker named = target" → forbidden. Always confirm.
• Picking direction from gut feel ("SNPS is bigger so it must be the acquirer") → forbidden. Use the 8-K filing or ask the user.
• Continuing past Step 0 with the direction still ambiguous → the entire report will be wrong; stop and resolve first.

Step 1 — Pull deal-defining filings

For each side (target + acquirer):

1. Pull recent 8-Ks from EDGAR submissions JSON. Filter for Item 1.01 and Item 2.01 in the last 180 days.
2. Pull the S-4 (acquirer) if any stock consideration.
3. Pull the DEFM14A or PREM14A (target).
4. Pull any 425 communications from the last 60 days.
5. Save under oneoff/ma_<TARGET>_<ACQUIRER>/.

Read the DEFM14A (or PREM14A if final not yet filed) for the canonical disclosure of: consideration mix, implied value, exchange ratio (if stock), termination fee, expected close window, board's justification, financial advisors' fairness opinions, target-board projections, target-management projections.

Step 2 — Compute the deal economics

Build a structured deal-snapshot block:

target: <Ticker> (<Name>, <Exchange>)
acquirer: <Ticker> (<Name>, <Exchange>)
announcement_date: YYYY-MM-DD
consideration_mix:
  cash_per_share: $X
  stock_ratio: Y acquirer shares per target share
  cap_collar: <if any>
implied_value_per_share: $Z (= X + Y × acquirer current price)
deal_value_total: $W billion (= implied_value × target shares outstanding)
expected_close: YYYY-MM-DD (range or quarter)
termination_fee: $T million
walk_away_conditions:
  - <bulleted list>

Then the spread block:

target_current_price: $A (as of YYYY-MM-DD)
deal_implied_value: $Z
spread: B% (= (Z - A) / A)
days_to_expected_close: C
annualized_return: D% (= spread × 365 / C)

Step 3 — Build the milestone tracker

Standard milestone path for a US public-to-public deal:

For non-US deals, substitute jurisdiction-specific milestones (UK CMA Phase 1 / 2, UK Takeover Panel Day 60 deadline, etc.).

Step 4 — Build the break-risk map

Score each of five break-risk categories on a 0–10 scale (0 = no risk visible; 10 = active blocker):

Total break-risk score (0–50) → narrative paragraph naming the dominant risk and what would change the picture.

Step 5 — Probability range and triggers

State the probability of close as a range, not a point estimate, with explicit triggers for the bear / base / bull paths:

The probability range should sum to 100% and have a reasonable spread — single-point "92% likely" is overconfident; "70/20/10" is healthy for a mid-pipeline deal.

When zsxq broker notes handicap the deal (≥2 notes): date every institute view (filename -YYMMDD suffix = report date) and flag disagreement on close probability explicitly — institute / date / implied probability or view / core argument, each citing its /zsxq/pdf/<file_id>/<urlencoded-name> link. An institute revising its own handicap as milestones pass is a dated What-to-Watch event, not a silent overwrite (per § M Sell-side view evolution (卖方观点演变)).

Step 6 — Recent news scan (last 30 days)

Web-search the last 30 days for:

• Antitrust / regulatory news (use regulator + target + acquirer query)
• Activist-investor 13D filings (rare but material)
• ISS / Glass Lewis vote recommendations
• Macro headlines that could move the spread (rate moves, credit-market stress, sector regulatory action)

Cite each material news item inline with a date in the link title. Skip items without verifiable URLs.

Step 7 — Write the report

Save to reports/ma/<Target>_<Acquirer>_<YYYY-MM-DD>.md. Suggested section structure:

1. Deal Snapshot — target / acquirer / consideration / implied value / expected close (the YAML block from Step 2, rendered as a clean table).
2. Spread & Annualized Return — current spread math, including the as-of date for the target price.
3. Milestone Tracker — table from Step 3.
4. Break-Risk Map — five-category score from Step 4 + narrative.
5. Probability & Triggers — Step 5's range with named triggers.
6. Antitrust Detail — jurisdiction-by-jurisdiction status (US HSR / EU DG-COMP / SAMR / CMA / others).
7. Background & Strategic Rationale — 1-2 paragraphs on why the buyer wants the target.
8. Recent News — last-30-days items with inline citations.
9. What to Watch Next — the next 1–3 dated events that should move the probability range.
10. Data Used manifest (mandatory).
11. References — every URL cited.

Output format (mode C — mandatory blocks)

Every mode-C report must contain:

1. Deal Snapshot at the top — target, acquirer, consideration, implied value, announcement date, expected close.
2. Spread & Annualized Return with explicit as-of date.
3. Milestone Tracker as a status table.
4. Break-Risk Map with five scored dimensions.
5. Probability range (not a point estimate) with named triggers.
6. ## Data Used / 数据来源清单 manifest.
7. ## Guardrails for this tracking pass block.

Pre-save checklist (Mode C — run before every save, including in-place updates): all 7 mandatory blocks above present; direction confirmed per Step 0; spread math shows all three inputs with as-of date; every newly added URL HTTP-checked (including xs-macbook-air.local zsxq links); when ≥2 zsxq notes handicap the deal, every institute view dated + same-institute revisions and disagreements flagged per Step 5 / § M Sell-side view evolution (卖方观点演变); Further-viewing block present or explicit "nothing to visualize" line; verification log appended per the cross-cutting guardrail.

Data Used / 数据来源清单 (mandatory)

## Data Used / 数据来源清单

**Deal-defining filings**
- DEFM14A or PREM14A (filed YYYY-MM-DD); S-4 (filed YYYY-MM-DD); 8-K Item 1.01 (target, filed YYYY-MM-DD); 8-K Item 1.01 (acquirer, filed YYYY-MM-DD); 425 communications (last filed YYYY-MM-DD). Source: SEC EDGAR.

**Antitrust / regulatory**
- US HSR (filed YYYY-MM-DD, second-request status: <yes/no>). EU Commission case M.xxxxx (filed YYYY-MM-DD, phase: <I/II>). UK CMA case ID xxxx (status). SAMR concentration filing YYYY-MM-DD (status). Other: <list>. Source: regulator press releases + case dockets.

**Pricing**
- Target price as of YYYY-MM-DD HH:MM (source: yfinance). Acquirer price as of same. Implied deal value computed inline.

**Recent news (last 30 days)**
- <N> news items: Bloomberg / Reuters / FT / WSJ / industry-trade press — each with inline citation + date. Source: WebSearch.

**Stale notices / coverage gaps**
- <bulleted list — financing condition not yet met / disclosed; jurisdiction's regulatory status not yet announced; or "none">.
- E.g.: "China SAMR concentration filing date not publicly disclosed; assumed accepted Q2 2026 based on standard timeline."

Guardrails (mode C)

• Never silently assume which side is target and which is acquirer. "SNPS-ANSS" reads either direction; getting it wrong corrupts the entire deal-snapshot block. Use explicit user phrasing ("X is acquiring Y"), or EDGAR 8-K Item 1.01 auto-detection when only one ticker is given, or ask the user before proceeding when two tickers are given ambiguously. The filename convention <Target>_<Acquirer> is a consequence of the resolved direction, not an input-format assumption. See Step 0.
• Do not state a deal closing date the company hasn't stated. "Expected by end of 2Q26" must trace to a specific filing or press release. Otherwise write "Expected close: per company disclosure, 'sometime in 2026'" and quote the exact language.
• Do not compute a spread without all three inputs disclosed. Deal-implied value + target current price (with timestamp) + days to assumed close — show all three.
• Do not skip a break-risk category just because it's not currently visible. "Antitrust: clear so far" is a valid score (1–2/10), but "not addressed" is not.
• Do not source antitrust outcomes to news articles when the actual regulator decision is public — cite the regulator's docket / press release at a specific URL.
• Do not promise a deal close. Probability is a range with named triggers; "this deal will close" is overconfidence regardless of how clean the picture looks.
• Do not confuse deal consideration with fair value. Deal consideration is what the buyer is paying; fair value is the target's standalone DCF. They are different.
• Do not silently drop a pending shareholder vote. Even on routine deals, the vote is a binary checkpoint — surface it in the milestone tracker.
• Do not write speculation about acquirer's next move post-close unless it's in a filing. "We expect SNPS to flip ANSS' physics-simulation IP into its EDA stack" is interpretation, not disclosure — label as *Analyst view:*.
• Do not run destructive SQL against db/*.db. Read-only only. See CLAUDE.md § "Database Safety".

Supplementary deliverables (mode C)

• M&A filings cache: oneoff/ma_<TARGET>_<ACQUIRER>/ (S-4, DEFM14A, 425 PDFs/HTML).
• Charts (optional): reports/charts/ma_<TARGET>_<ACQUIRER>_*.png (e.g. target-price chart vs deal-implied value; spread over time).

Update-in-place rule (mode C)

One English file and one Chinese file per ordered (target, acquirer) tuple. Each tracking pass updates the date suffix in the filename to the new pass date. Older dated copies may exist if this is a multi-month tracking effort; treat them as a snapshot history.

Event positioning lens — option / VIX / spread context (Modes A & B)

Bull/bear thresholds on a macro print or earnings release are only half the picture. The other half is what's already priced — if the market enters the print already short, a hot CPI delivers a smaller move than the same print into a long-positioned tape. This section codifies how to use option pricing, vol-surface signals, and credit spreads to translate the indicator-level bull/bear thresholds into actionable trade context.

Apply this lens to Mode A briefs (day-of) and Mode B previews (week-ahead). Skip it for Mode C (single-deal M&A) — the spread math there is direct, not vol-based.

The 4-layer framework

For every binary event (macro print, earnings, FOMC, vote outcome), ask in order:

1. What's priced? — what does the consensus position imply? Tells you the size of the surprise needed to move the tape.
2. What's the implied move? — what does the option market literally pay for movement on the print day?
3. What's the directional bias? — skew + put-call ratio + 25-delta-put-vs-call IV; is downside already paid for, or is it fresh?
4. Cross-check — do credit spreads, MOVE, and equity vol agree on the risk premium? If one disagrees, the dissenter is usually wrong by 1–2 standard deviations.

Toolkit

Per-indicator playbook

Each macro indicator has a primary pricing signal and a "too high" tell. Use the primary tool to gauge the size of the surprise needed; the "too high" tell flags when the market has over-positioned for one outcome.

Truncated - read the full file at https://github.com/dadachundan/financial_agent/blob/e38112fa49dd368b650012b7f28627e3826183ee/.claude/skills/catalyst-calendar/SKILL.md.