chore: 1.1.0 release

skills/open-kimi-ppt/reference/design_system/finance/black-gold-ledger/design.md

@@ -0,0 +1,94 @@
+# Black Gold Ledger · finance STYLE DESIGN SYSTEM
+
+## PART A — finance Baseline (Universal Floor)
+
+Institutional-grade investment research presentations, benchmarked against top-tier fund research reports, due diligence, and strategy outlooks. Chart-heavy, text on the dense side, more report than consulting: Excel-style financial model tables carry the argument; every number is traceable, every assumption made explicit, every conclusion executable.
+
+Content and layout:
+1. Thesis-driven structure: deal decks run "summary → highlights → target → financial diagnosis → valuation & returns → risk mitigation → decision ask"; research decks run "stance → evidence → scenarios → allocation advice". The ending must land on an explicit decision or stance.
+2. Tables are the protagonist: valuation ranges, comparable companies, the three financial statements, scenario modeling, sensitivity grids — all under strict row/column discipline (assumption columns, driver rows, check rows); charts serve the numbers, never decorate.
+3. High but ordered density: multi-column layouts with clear hierarchy; each page answers one investment-committee question; density comes from evidence modules, not walls of text.
+
+Visual style: restrained, authoritative, compliance-clean; no lifestyle photography, decorative icons, or marketing superlatives. Serif is allowed for titles when gravitas is needed; table body text, chart labels, and all figures use sans-serif, with numbers column-aligned. A conservative institutional palette: deep structural colors plus one restrained accent, reserved for key figures and recommended actions; positive/negative semantic colors fixed, expressing financial meaning only. Fine table rules, precise vector chart lines; no 3D or shadows; gradients are not part of the standard chart language.
+
+Structure and discipline: every body-page title must be a complete investment assertion, with a one-line subtitle stating scope and how to read it; every number must carry unit, as-of date, and measurement basis — estimates must be flagged, no bare numbers; returns are given only as scenario ranges (bear/base/bull, with probabilities or assumptions), never a single-point IRR, and every valuation conclusion ships with a sensitivity table; assumptions are boxed and made explicit, placed on the page whose conclusion depends on them; risks and mitigations appear in pairs (risk → probability/impact → mitigation → owner); every valuation page must have an anchor (comparable companies, precedent transactions, or DCF cross-check); parallel evidence runs in 2–3 equal-width columns, numbered arguments carrying one bolded claim line plus one proving sentence; chart labels sit on the marks, forecast segments use dashed lines, key drivers are named inside the chart — never rely on the legend alone; the deck closes on a decision or stance, with next actions and a timeline — never end on a thank-you page.
+
+Default baseline (PART B may override): key metrics are set in a size larger than body text, adjacent to their units and basis; at least one class of high-density financial table is kept, with right-aligned numbers, thousands separators, and a unified negative-number convention; positive/negative semantic colors fixed across the deck; forecast vs. historical visually distinguishable and labeled by basis; each page carries source, basis, and time range in small gray footer text; a single temperament per deck — choose one. Headline read-through test: reading only the page titles across the deck should tell the complete story. Density anchor: per-page minimums for evidence modules and charts follow the PART B density baseline.
+
+Global prohibitions (mandatory): no default cards (information containers such as status lights, assumption boxes, and verdict bands do not count as cards); no unearned even splits (three/four-way splits and 2×2 grids must be justified by content); no unmotivated AI color schemes (blue-purple gradients, cyan-purple neon, glass cards, glowing borders); no elements that clash with the overall style; chart backgrounds transparent or same as the page background (except reversed-out on dark); no large empty blocks in the body area — planned regions must be filled with real content.
+
+## PART B — Signature System
+
+One-line style signature: financial-periodical order in black, white, and gray — white evidence pages carry verifiable numbers via axis-free charts and hairline tables, dark pages hold only judgments and decisions, a single gold anchors current values and key thresholds, and no stray hues appear anywhere in the deck.
+
+【Color Palette】
+All values below are approximations sampled from the reference images; a single gold anchor, grayscale steps carrying data, and two background depths with distinct duties.
+- White background #FFFFFF (approx.): about 3/4 of pages, carrying verifiable data and detail.
+- Dark background #151515 (approx.): charcoal, not pure black; used only for summary and closing-action pages, about 1/4.
+- Ink black #0A0A0A (approx.): titles on white pages, top rule, line series, black full-width bands, and the strongest text.
+- Warm dark gray #3D3C3A (approx.): body text, primary-series bars, key in-table figures.
+- Secondary gray #606060 (approx.): subtitle basis line, chart captions, units.
+- Supporting gray #A7A7A7 (approx.): secondary series, column labels, auxiliary text on dark.
+- Footnote gray #B6B6B6 (approx.): sources, page numbers, and other weakest-tier information.
+- Light gray #D6D6D6 (approx.): historical/background series, table hairlines, caption text on dark.
+- Panel gray #F0F0F0 (approx.): watermark-sized type, conclusion bands, light reversed cards on dark.
+- Track gray #F3F3F3 (approx.): pale track beds for progress and composition bars.
+- Primary gold #D6A000 (approx.): the single strong accent anchor — KPI big numbers, current endpoints, threshold capsules, the short gold cover line, key horizontal bars; under 3% of any single page's area, one gold focus per page, no light/dark double gold.
+- Copper brown #A86038 (approx.): the only auxiliary semantic color, reserved for exit/risk-related amounts, low frequency.
+- Structural lines: on white pages #0A0A0A (approx.) sets page-level boundaries, #D6D6D6 / #E9E9E9 (approx.) separate rows and modules; on dark pages the divider is #302F2D (approx.).
+- No red/green up-down system (overrides the PART A baseline): gold marks current and key values, copper brown marks exits/risk, and all other data recedes into neutral gray steps.
+
+【Layout Skeleton】
+Source facts: 16:9 canvas (source images 1467×825 px, generated at 960×540 pt); left/right margins about 5.5% of canvas width (~52 pt), content span about 89%. The grid resolves to 12 columns, combined per content into thirds, roughly 2:1, roughly 1:1, or five-column tables, overall asymmetric; no logo, no breadcrumbs — the English masthead on the cover carries the publication identity.
+- White analysis pages: thin black top rule (about 5% from the top) → conclusory title → gray basis subtitle → light-gray divider → body (starting around 1/5 of page height) → footer with basis/source at left and "n / N" page number at right.
+- Dark pages: letterspaced small English label → white conclusory title → one to two lines of explanation → numbered judgments or full-width light cards → bottom judgment/decision line.
+- Cover/closing: masthead and date at top left and right → short gold line → two-line large title → abstract → KPI big-number row → oversized watermark at right → black basis band across the bottom.
+- Reading path: title assertion → main chart/evidence area → explanation column or KPIs → bottom judgment and basis line; the gold focus closes the attention loop last.
+
+【Typography】
+Modern sans-serif for both CJK and Latin (Source Han Sans temperament). Title weight 700–800, section titles 600–700, body 400, footnotes lighter. Taking body text as 1x: standard page titles about 2.1–2.4x, cover title about 3x, KPI big numbers about 2.8–3.5x, chart labels about 0.8–0.9x, footnotes about 0.7x (reference at 960×540: body 9–10 pt, page titles 21 pt, cover 36 pt). Big numbers share the line with their units, units shrunk to 50%–70%, basis on the next line; numbers bolded, tight against sign and unit. Small English labels in all caps, light weight, tracking 0.20–0.28em — a financial-periodical header signature, used only on the cover and dark pages.
+
+【Chart Language】
+- ⭐ Bar–line combo (amount bars + count line): historical bars light gray, current-period bars dark charcoal, the line in ink black; bar values on top, point values above nodes; key changes called out with gold capsules.
+- ⭐ Full-width time-series line: ink-black ~2 pt line with circular nodes, the current endpoint enlarged and switched to gold; only a ~1 px baseline remains — no y-axis, grid, or frame.
+- ⭐ Horizontal bar groups (ranking/comparison): dark charcoal for the focus item, mid/light grays for the rest and for history; a key series may run entirely in gold; values labeled directly at bar ends, multiple bar groups sharing row baselines.
+- ⭐ Grayscale composition bar (100% segments): dark → mid → light segments, reversed-white percentages inside, name and amount annotated in two lines below.
+- Dot plot (aligned comparison, low frequency): shares row baselines with horizontal bars; focus-row dots in gold or dark charcoal.
+- Step bars (staged progression, select cases): light → mid → dark progression, expected range values flagged in gold.
+- In-table progress bars: dark charcoal bars on pale gray tracks embedded in vertical-line-free tables, expressing structural share.
+- KPI big-number column: right-hand vertical rail or horizontal row on the cover; gold goes only to the one or two most critical numbers.
+- Universal discipline: all data labels placed directly (bar tops / bar ends / beside nodes), never legend-dependent; tables are ruled primarily by horizontal lines — no zebra striping, no dense vertical lines; chart backgrounds match the page.
+
+【Signature Components】
+- Thin black top rule: fixed header fixture of white analysis pages, ~2 pt full width, the identity line of evidence pages.
+- Short thick gold line: about 46×4 pt, set above the cover/closing main title, one per special page only.
+- Letterspaced small English label (eyebrow): all caps, tracking 0.20–0.28em, reserved for the cover masthead and dark-page headers.
+- Gold capsule: fully rounded gold chip with ink text, carrying only key changes, time cross-sections, or trigger thresholds — at most one per page.
+- Right-hand KPI rail: a vertical line sets it off from the main area, big numbers alternating with gray captions; gold/white on dark pages, gold/ink on white pages.
+- Oversized light-gray watermark type: panel gray, more than ten times body size, holding a year or institution initials at the right of cover/closing pages, low frequency.
+- Bottom black basis band: full-width black bar at the foot of special white pages, small white text carrying fund information and risk disclosures.
+- Light-gray verdict band: full-width panel-gray strip at the foot of white pages, closing the page with one bolded judgment.
+- Large light-gray numerals: dark key-point pages only, numbered from 01, paired with bolded white assertions.
+- Footer "basis + page number" pair: small gray basis/source at left, n / N at right — present on every page of the deck; legends use small squares only, data points small circles only, never decorative.
+
+【Prohibited】
+- No colored chart series or second accent: apart from gold (and low-frequency copper brown), all data series run in gray steps.
+- No red/green up-down pairs: positive/negative semantics are not carried by red and green (overrides the PART A baseline).
+- No pies, donuts, radars, stacked areas, 3D, shadows, gradients, glows, or default Office chart styles.
+- No y-axes, gridlines, or chart frames — axes keep only the baseline; no chart that can only be read via its legend.
+- No complex data grids on dark pages: dark pages carry only stage judgments, executive summaries, and decision asks.
+- No sprinkling dark backgrounds across analysis pages: dark pages are about 1/4, fixed at the summary and closing-action pages.
+- No photos, illustrations, colored icons, large textures, or logo dependence.
+- No rounded-card walls or evenly split card grids: the light full-width card exists only as the dark action-page form.
+- No centered layouts or symmetric display titles: everything left-aligned, asymmetric columns.
+- No large-area gold fills or full pages of gold type (<3% per page, one focus per page).
+
+【Slide Types and Layouts】
+Cover snapshot | the deck's single white front page | left ~58% holds the masthead, short gold line, two-line large title, two-line abstract, and three KPIs; oversized watermark at right; black basis band at bottom | ~6 information blocks, the most whitespace in the deck | not suited to detailed data.
+Dark key-points page | executive summary or next-quarter action list | small label + white conclusory title + caption line; left ~2/3 numbered judgments, right ~1/3 KPI rail; or three full-width light cards organized as "observation — gold-capsule threshold — trigger action" | 3 judgments + 3–4 KPIs, or 3 threshold cards plus default actions | full-width overall judgment / decision ask at bottom | not suited to complex charts and grids.
+Trend data page | time series of totals, revenue, etc. | one main chart in the upper half (bar–line combo or full-width line), lower half in three equal explanation columns or a dual module of "composition bar + step bars" | 1 main chart + 2–4 explanation blocks | gold marks only the current bar, current endpoint, and key capsules | not suited to parallel multi-topic content.
+Table matrix page | segment/portfolio breakdown | five-column, four-row vertical-line-free table: name, amount big number, embedded progress bar, representative items, through-line; light-gray hairlines between rows, ink line before the last row, total row + meaning row at bottom | highest density in the deck (~20 information units), noise-reduced via generous row height and gray steps | not suited to long-form narrative.
+Comparison ranking page | region/project/approach comparison | horizontal bar groups occupy the left ~2/3, a KPI quick view or dot plot the right 1/3, sharing row baselines; two same-scale bar groups may sit side by side below, closed by a light-gray verdict band | 3–5 bar rows + ~3 KPIs | dual emphasis in gold or dark charcoal goes only to the leader | not suited to time-series narrative.
+
+【Density Baseline】
+Body pages carry 3–7 information blocks; the standard combination is "one main chart + 2–4 explanation or KPI blocks"; text blocks run 1–2 lines, roughly 20–55 words/characters equivalent, footnotes a single line. Charts directly label 3–8 data points — small samples fully labeled, long series labeled only at first/last and key points; table pages allow ~20 information units, noise-reduced with generous row height, hairlines, and gray steps. Whitespace concentrates in chart backgrounds, inter-column gutters, and the title-to-body transition band — never spread evenly. Dark pages carry fewer blocks, building authority through reversed contrast rather than volume. One gold focus per page; dark pages ~1/4 of the deck; the deck-wide rhythm is "white cover — dark summary — a run of white evidence — dark action close".

skills/open-kimi-ppt/reference/design_system/finance/ebony-ledger/design.md

@@ -0,0 +1,72 @@
+# Ebony Ledger · Finance STYLE DESIGN SYSTEM
+
+**Full creative commitment**: apply your complete creativity and imagination, treating every page as an independent, complete design work worthy of scrutiny. Structure, composition, charts, imagery, and visual narrative must be content-driven — never fall back on default layouts.
+
+**No AI slop**: no low-judgment, templated, decoration-stacked AI default output, including unjustified card arrays, evenly split grids, gratuitous blue-purple gradients or glassmorphism, icon walls, hollow oversized type, repeated skeletons, and visual gimmicks unrelated to the content. Every page must show clear design judgment with a basis in the content.
+
+**Landscape delivery**: the final presentation is uniformly 16:9 landscape. Portrait or non-16:9 source material serves only as style observation and must be recomposed into 16:9 landscape — the original canvas ratio must not be carried over directly.
+
+**Layout coordination**: style.md governs the final style rules (fonts / colors / components / chart language / prohibitions); the "layout / visual task" line that query.md gives for each page governs that page's content structure, primary anchor, and reading order — both are binding execution specs. Layout intent follows the query task line first; final visuals follow style.md.
+### Global Prohibitions
+
+1. **No default cards**: unless the user explicitly requests it, never use rounded rectangles or rectangular cards to build hierarchy or alignment — containers that serve an information function (status lights, assumption boxes, quote frames) do not count as cards. Rules, whitespace, and differences in type size and weight are the better tools.
+2. **No evenly split compositions**: unless no other layout is truly available, do not default to three-way splits, four-way splits, or 2×2 matrices — even splits must be justified by content structure (three scenarios naturally take three columns; frameworks explicitly named by the content are exempt).
+3. **No unjustified AI color schemes**: institutional palettes specified in the content are exempt; what is banned are bland, common, unjustified default colors — blue-purple gradients, cyan-purple neon, rainbow flares, glass cards, glowing borders.
+4. **No elements that clash with the overall style**: nothing outside the style system may appear — for example, rounded icons or rounded rectangles in a sharp style.
+5. **Transparent chart backgrounds**: transparent or the page background color, except reverse-white treatments on dark grounds.
+6. **Avoid large blank areas**: keep reasonable density in the body area; padding out or diluting content into sparse pages is treated as a defect. Whitespace for breathing, grouping, and rhythm is allowed — form follows content; never stuff material in just to fill the page.
+
+## PART A — Category Baseline (universal floor)
+
+Institutional-grade investment research presentations, benchmarked against top-tier fund research reports, due diligence, and strategy outlooks (bulge-bracket bank and sovereign wealth fund standards). Chart-heavy, text-dense, more report-like than consulting decks: arguments are carried by Excel-style financial model tables (three-statement models, scenario analyses, sensitivity grids). Every number traceable, every assumption explicit, every conclusion actionable.
+
+Content and layout:
+1. Thesis-driven structure: deal decks run "Summary → Investment highlights → Target/Market → Business plan/Financial diagnostics → Valuation and returns → Risks and mitigants → Decision request"; research outlooks run "Core stance → Evidence → Scenario analysis → Allocation recommendation". The ending must land on an explicit decision or stance.
+2. Tables are the protagonist: valuation ranges, comparable companies, three financial statements, scenario analyses, sensitivity grids, with Excel financial-model row/column discipline (assumption columns, driver rows, and check rows clearly distinguished); charts serve the numbers (football-field valuation charts, IRR bridges, waterfall attributions, scenario fan charts) and are never decorative.
+3. High but ordered density: multi-column layouts with clear information hierarchy; each page answers one investment-committee question; density comes from evidence modules, not walls of text.
+
+Visual style:
+1. Aesthetic: restrained, authoritative, compliance-clean. No lifestyle photography, no decorative icons, no marketing superlatives.
+2. Typography: use serif for titles when institutional gravitas is needed; sans serif for table body text, chart labels, and all numerals; numbers are always column-aligned.
+3. Color: a conservative institutional palette — navy/charcoal structural colors, one restrained accent (e.g., gold or signal blue) reserved for key figures and recommended actions; positive/negative signs use one consistent red-green pair, expressing financial meaning only.
+4. Graphics: thin table rules, precise vector chart lines, fine connector lines on bridge charts; no 3D, no shadows, and gradient fills are not a routine chart language.
+
+Structure and discipline (distilled from the artifacts we consider benchmarks):
+1. Every body-page title must be a complete investment assertion (e.g., "This asset will reprice on capacity scarcity" — not "Market overview"), with a one-line subtitle stating scope and how to read the page.
+2. Every number carries unit, as-of date, and basis; estimates must be labeled (illustrative / internal estimate); no bare numbers.
+3. Returns are given only as scenario ranges (bear/base/bull, with probabilities or driving assumptions) — never a single-point IRR; any valuation conclusion must be paired with a sensitivity table.
+4. Assumptions must be made explicit in boxes, placed on the page where the conclusion depends on them; hidden assumptions are treated as defects.
+5. Risks and mitigants appear in pairs (risk → probability/impact → mitigant → owner/status); a page with risks but no mitigants is a defect.
+6. Valuation pages must be anchored: comparable companies, precedent transactions, or a DCF cross-check — a multiple without a reference set is not evidence.
+7. Parallel evidence runs in 2–3 equal-width columns; numbered arguments each carry a one-line bolded claim plus one short proof sentence.
+8. Chart annotation discipline: data labels attached to marks, bridge-chart connector blocks labeled block by block, forecast segments dashed, key drivers named inside the chart; charts may not rely on legends alone.
+9. The ending must land on a decision or stance: items submitted for decision (approve / reject / conditionally approve) or an explicit allocation stance (overweight / underweight / range), with a to-do list and timeline — never close with a thank-you page.
+
+Default baseline (distilled from the layout skill; PART B may override): numbers first — key metrics (amounts / growth rates / multiples) must be set larger than body text and presented standalone, immediately adjacent to unit and basis; keep at least one class of high-density financial table (three statements / comps / sensitivity), with right-aligned numbers, thousands separators, and negative-number conventions unified deck-wide; positive/negative semantic colors fixed — up/positive and down/negative each use one fixed color that never changes across the deck; forecast and history must be visually distinguishable (dashes / light tints / hatched bands) and labeled with the basis; every page carries source, basis, and time range in small gray type at the page bottom — numbers without sources are not allowed in the main visual area; a single temperament per deck (dark navy/charcoal with metallic accents OR traditional white ground with blue-gray) — pick one per deck.
+
+Density anchor: each style's specific floor (evidence modules per page, chart count, annotation habits) follows the 【Density Baseline】 reverse-engineered from that style's source material — this baseline imposes no uniform numbers.
+
+
+## PART B — Signature System (this template's signature; where it conflicts with PART A, this section governs)
+
+One-line style signature: a calm institutional-annual temperament — pure white ground with a black skeleton, teal-green data, and terracotta conclusions; medium-high density that retains magazine-like breathing room.
+
+【Color Palette】 Background #FFFFFF; primary body text and headings #000000 — black carries only text, baselines, and the top bar. Primary accent #4C9F8B for current values, main scenarios, or winning items, occupying no more than one-third of any single chart. Supporting #1B5E5C for secondary series or starting values. Conclusion color #A95228 reserved for multipliers, arrows, and short vertical ticks — never large color blocks. Structural gray #CACED7 for historical values; light gray #E4E4E4 for annotation plates and range bands. Chart series pick from #4C9F8B, #1B5E5C, #CACED7, #C39D78; when positive/negative encoding is truly needed, positive is #4C9F8B and negative is #C39D78 — #A95228 does not mean negative. At most two colored hue families per page; the iron rule is that the accent must never blanket the page.
+
+【Layout Skeleton】 Source facts: 612×792 US Letter portrait, 17:22; wide, stable margins on all four sides; a black horizontal rule placed inset at the top of the page, about 80% of page width. Body is single-column or 48:48 two-column with narrow, stable gutters; reading runs top-to-bottom, left-to-right. No navigation or breadcrumbs; the footer pairs a short running title at left with a thin vertical rule and page number at right. Overriding PART A: section pages may use the theme name as the main title, with the assertion demoted to a weaker subtitle; continuation pages may omit the page title and open with the chart head instead. Cover only: the top rule's left end is indented to about 31% of page width, the title uses two levels of serif, and the lower half page carries a key-point list marked by terracotta vertical ticks. 16:9 migration: unify finally to 16:9 landscape, keep the top bar, left alignment, and two columns, and recompose vertical chart groups into 5:7 or 6:6; the title area takes 14%–17%, the body area about 76%, the footer about 5%; long passages must be restructured.
+
+【Typography】 Source facts: headings use the embedded serif SanomatTT-Light, with key numerals occasionally in its Medium/Semibold; body, legends, axis labels, and footers use the sans serif GuardianSansTT-Light, with labels in Medium/Regular. Page titles sit clearly above paragraph heads and chart titles but stay restrained overall; chart titles and paragraph heads are only slightly stronger than body. Key numerals may be raised to title level or slightly above when a conclusion genuinely needs focus; footnotes, axis labels, and footers recede noticeably. Conclusion numerals use serif; ordinary data labels use sans serif. Render with {latin: Unna, ea: 思源宋体} (serif titles and conclusion numerals) and {latin: Liter, ea: MiSans} (sans body/labels). In 16:9 landscape, keep the light weights and this relative hierarchy — do not carry over the source's absolute sizes or ratios.
+
+【Chart Language】 Actually used: horizontal ranking bars, standard/positive-negative/stacked columns, line charts, donut charts, big-vs-small circle comparisons, benchmark dashes, and range bands; tables are not used — when financial tables are required, inherit PART A. ⭐ Historical bars are gray; current or main-scenario bars are teal-green; discrete bars carry direct value labels. ⭐ Change conclusions use large terracotta multipliers with black slanted arrows or brackets — at most once per chart. ⭐ Line charts label only endpoints, turning points, or comparison conclusions. ⭐ Forecast increments use teal-green dashed outlines; recession periods or explanatory zones use light-gray rectangular bands. ⭐ Parallel evidence may form a 2×2 small-chart matrix, each chart keeping its own chart number, serif chart title, and basis line.
+
+【Signature Components】 Thick black top bar: the header anchor, once per page, hugging the top margin only. Three-layer chart head: sans-serif medium-black chart number + serif chart title + small basis line — once per chart. Terracotta conclusion multiplier: extracts the magnitude of change, at most once per chart and twice per page, placed right next to the arrow or bracket. Light-gray evidence plate: carries benchmarks or summaries, 0–2 per page, hugging chart edges only. Minimal footer: short running title and page number at opposite ends, once per page.
+
+【Prohibited】 No dark full-bleed pages, dark section pages, or large teal-green backgrounds; no replacing teal-green with navy or bright blue; no heavy sans-serif page titles; no treating #A95228 as a negative-warning color; no colored blocks without data meaning and no more than four series colors. Use photos / illustrations cautiously, never as backgrounds.
+
+【Slide Types and Layouts】 Single-column argument + main chart ｜ one thesis with one core piece of evidence ｜ upper argument ~30%, middle main chart ~45%, lower explanation ~25% ｜ 3–4 blocks ｜ top bar, three-layer chart head, current value in teal-green ｜ not suited to four parallel conclusions.
+Staggered two-column text-image ｜ two sets of evidence explaining each other ｜ ~48:48 two columns, vertical offset allowed ｜ 4–6 blocks ｜ terracotta multipliers, light-gray evidence plates ｜ not suited to continuous long text.
+Multi-chart evidence board ｜ cross-validating one conclusion from multiple angles ｜ title and intro ~20%, body is a 2×2 or one-large-two-small chart group ｜ 4–6 blocks, 2–4 charts ｜ independent chart heads, direct value labels ｜ not suited to a single complex causal chain.
+Section opening page ｜ opening an asset or theme section ｜ title ~12%, stance and background ~28%, evidence area ~60% ｜ 3–5 blocks ｜ top bar, large serif title, 1–2 charts ｜ not suited to slogans only.
+Pure-argument closing page (occasional) ｜ summarizing a stance or disclosing boundaries ｜ single column of 3–4 paragraphs, ~80% wide ｜ 2–4 blocks ｜ top bar and footer only ｜ not suited to number-dense topics.
+
+【Density Baseline】 Source body pages carry about 230–438 English words, 3–6 evidence modules, 1–4 charts; narrative pages 300–440 words with 0–2 charts; multi-chart pages 230–350 words with 2–4 charts. For 16:9 multi-chart pages, recommend 3–5 modules, 2–4 charts, and 1–2 explanation blocks of 35–70 words/characters equivalent each; narrative pages use 2–4 text blocks of 60–100 words/characters equivalent each, with 0–1 chart. Discrete bars each carry a value label; line charts mark only key points. Place whitespace after titles, between chart groups, and above the footer. No uniform hard floor across page types.

skills/open-kimi-ppt/reference/design_system/finance/honey-orange-memo/design.md

@@ -0,0 +1,74 @@
+# Honey Orange Memo · Finance STYLE DESIGN SYSTEM
+
+## PART A — Category Baseline (universal floor)
+
+Institutional-grade investment research presentations, benchmarked against top-tier fund research reports, due diligence, and strategy outlooks (bulge-bracket bank and sovereign wealth fund standards). Chart-heavy, text-dense, more report-like than consulting decks: arguments are carried by Excel-style financial model tables (three-statement models, scenario analyses, sensitivity grids). Every number traceable, every assumption explicit, every conclusion actionable.
+
+Content and layout:
+1. Thesis-driven structure: deal decks run "Summary → Investment highlights → Target/Market → Business plan/Financial diagnostics → Valuation and returns → Risks and mitigants → Decision request"; research outlooks run "Core stance → Evidence → Scenario analysis → Allocation recommendation". The ending must land on an explicit decision or stance.
+2. Tables are the protagonist: valuation ranges, comparable companies, three financial statements, scenario analyses, sensitivity grids, with Excel financial-model row/column discipline (assumption columns, driver rows, and check rows clearly distinguished); charts serve the numbers (football-field valuation charts, IRR bridges, waterfall attributions, scenario fan charts) and are never decorative.
+3. High but ordered density: multi-column layouts with clear information hierarchy; each page answers one investment-committee question; density comes from evidence modules, not walls of text.
+
+Visual style:
+1. Aesthetic: restrained, authoritative, compliance-clean. No lifestyle photography, no decorative icons, no marketing superlatives.
+2. Typography: use serif for titles when institutional gravitas is needed; sans serif for table body text, chart labels, and all numerals; numbers are always column-aligned.
+3. Color: a conservative institutional palette — navy/charcoal structural colors, one restrained accent (e.g., gold or signal blue) reserved for key figures and recommended actions; positive/negative signs use one consistent red-green pair, expressing financial meaning only.
+4. Graphics: thin table rules, precise vector chart lines, fine connector lines on bridge charts; no 3D, no shadows, and gradient fills are not a routine chart language.
+
+Structure and discipline (distilled from the artifacts we consider benchmarks):
+1. Every body-page title must be a complete investment assertion (e.g., "This asset will reprice on capacity scarcity" — not "Market overview"), with a one-line subtitle stating scope and how to read the page.
+2. Every number carries unit, as-of date, and basis; estimates must be labeled (illustrative / internal estimate); no bare numbers.
+3. Returns are given only as scenario ranges (bear/base/bull, with probabilities or driving assumptions) — never a single-point IRR; any valuation conclusion must be paired with a sensitivity table.
+4. Assumptions must be made explicit in boxes, placed on the page where the conclusion depends on them; hidden assumptions are treated as defects.
+5. Risks and mitigants appear in pairs (risk → probability/impact → mitigant → owner/status); a page with risks but no mitigants is a defect.
+6. Valuation pages must be anchored: comparable companies, precedent transactions, or a DCF cross-check — a multiple without a reference set is not evidence.
+7. Parallel evidence runs in 2–3 equal-width columns; numbered arguments each carry a one-line bolded claim plus one short proof sentence.
+8. Chart annotation discipline: data labels attached to marks, bridge-chart connector blocks labeled block by block, forecast segments dashed, key drivers named inside the chart; charts may not rely on legends alone.
+9. The ending must land on a decision or stance: items submitted for decision (approve / reject / conditionally approve) or an explicit allocation stance (overweight / underweight / range), with a to-do list and timeline — never close with a thank-you page.
+
+Default baseline (distilled from the layout skill; PART B may override): numbers first — key metrics (amounts / growth rates / multiples) must be set larger than body text and presented standalone, immediately adjacent to unit and basis; keep at least one class of high-density financial table (three statements / comps / sensitivity), with right-aligned numbers, thousands separators, and negative-number conventions unified deck-wide; positive/negative semantic colors fixed — up/positive and down/negative each use one fixed color that never changes across the deck; forecast and history must be visually distinguishable (dashes / light tints / hatched bands) and labeled with the basis; every page carries source, basis, and time range in small gray type at the page bottom — numbers without sources are not allowed in the main visual area; a single temperament per deck (dark navy/charcoal with metallic accents OR traditional white ground with blue-gray) — pick one per deck.
+
+**Title read-through test**: reading only the page titles across the deck should tell the complete story; if the titles do not read through, the structure has not been established.
+
+Density anchor: each style's specific floor (evidence modules per page, chart count, annotation habits) follows the 【Density Baseline】 reverse-engineered from that style's source material — this baseline imposes no uniform numbers.
+
+### Global Prohibitions (AI default routines · mandatory)
+
+1. **No default cards**: unless the user explicitly requests it, never use rounded rectangles or rectangular cards to build hierarchy or alignment — containers that serve an information function (status lights, assumption boxes, quote frames) do not count as cards. Rules, whitespace, and differences in type size and weight are the better tools.
+2. **No evenly split compositions**: unless no other layout is truly available, do not default to three-way splits, four-way splits, or 2×2 matrices — even splits must be justified by content structure (three scenarios naturally take three columns; frameworks explicitly named by the content are exempt).
+3. **No unjustified AI color schemes**: institutional palettes specified in the content are exempt; what is banned are bland, common, unjustified default colors — blue-purple gradients, cyan-purple neon, rainbow flares, glass cards, glowing borders.
+4. **No elements that clash with the overall style**: nothing outside the style system may appear — for example, rounded icons or rounded rectangles in a sharp style.
+5. **Transparent chart backgrounds**: transparent or the page background color, except reverse-white treatments on dark grounds.
+6. **No regional blank space**: the body content area keeps no swaths of blank space; any planned region must be filled with real content — failing to fill it is the most severe form of sloppiness.
+
+## PART B — Signature System (this template's signature; where it conflicts with PART A, this section governs)
+
+One-line style signature: an investment-committee, report-style financial tone; warm-white ground, deep navy skeleton, bright orange anchors, cool blue and light peach as support — high information density with clear layering.
+
+【Color Palette】
+Background approx. #FAF8F6, with white #FFFFFF reserved for tables or note areas; primary text #4D4D55; titles and primary accent #FF6600; left structural rail approx. #FF5703. Deep structural color #172856 for active navigation segments, table headers, side labels, and argument bands; supporting #25408B and #579DB7 for series and comparisons; #FFE0CC and #96C2D2 for note blocks or row groups; #A6A6A6/#D9D9D9 as separators only. Chart series map orange → deep blue → cyan-blue; forecast segments switch to light tints or textures; positive/negative fixed at #008000 / #C00000, expressing financial direction only. Hold 4–6 meaningful hues per page; orange must not blanket the body.
+
+【Layout Skeleton】
+Source facts: 960×540 pt, 16:9. Nearly every page has a full-height left orange rail; standard pages carry a five-segment chevron navigation at top — current segment deep blue, the rest gray. The title area sits upper-left: bold orange title ~28 pt, gray subtitle ~16 pt, body starting at ~110 pt height; the body favors left chart/table + right commentary, an upper conclusion band over a lower evidence band, or three columns. The footer keeps only source/notes and page number. Section pages (5 in total) carry only a large orange number + deep blue title; the cover is a special case — full-bleed dark photography, deep navy overlay, white title, and an orange date block — and is not reused in body pages.
+
+【Typography】
+Source uses the sans-serif Arial family: ArialMT for body, Arial-BoldMT for titles/emphasis, a little Helvetica-Bold in footers; no stable serif observed. Body 10.6–11 pt = 1x, subtitles 16 pt ≈ 1.45x, regular titles 28 pt ≈ 2.5x, section numbers/titles 44 pt ≈ 4x, key large numbers 16–32 pt ≈ 1.5–3x, footnotes 7–8 pt ≈ 0.65–0.75x. Titles, conclusion bands, and key figures use bold; CAGR, multiples, and units may use italics; English abbreviations and years do not switch fonts.
+
+【Chart Language】
+Actual types: historical/forecast stacked columns and lines, event-annotated price lines, driver bridge charts, sensitivity matrices, comparable-company tables, historical/forecast financial tables, risk matrices, process cycle diagrams. ⭐ Forecast segmentation: table headers grouped in gray/deep blue/orange, forecast cells with light tints or hatching, CAGR and key years labeled directly on charts. ⭐ Event line charts: thin orange frames hung directly off data points. ⭐ Bridge charts: drivers color-coded, terminal value as a tall orange bar, with an item-by-item explanation row below. ⭐ Sensitivity matrices: deep-blue row/column headers, the base case circled with a green dashed line or orange border. ⭐ Financial model tables: multi-level headers, with grouped rows and total rows distinguished by gray/orange/light-yellow fills.
+
+【Signature Components】
+Fixed components are the five-segment chevron navigation and the left orange rail; deep-blue horizontal title bands with white text, about 2–5 per page. Orange numbered circles / lettered circles attach to the left end of argument bands — no more than 6 per argument page. Deep-blue, orange, and light-blue side labels head risk/stage/status rows without pressing into the body. Light-orange or light-blue note blocks carry only assumptions or commentary, 1–3 per page. Cycle flywheels and dashed rounded process frames are occasional components, appearing only on process pages.
+
+【Prohibited】
+No flattening the chevron navigation into straight tabs, and no removing the left orange rail; no expanding orange into large background areas; no body photography or decorative illustration — photography lives only on the cover; no stuffing charts or long text into section pages; no rounded containers without an information function — a few dashed rounded frames on process pages are the only occasional exception.
+
+【Slide Types and Layouts】
+Argument–evidence band page ｜ one assertion with 3–6 proofs ｜ title area + horizontal deep-blue argument band, or 60/40 left-right ｜ 3–6 blocks ｜ numbered circles, title bands ｜ not suited to line-by-line financial detail.
+High-density chart–commentary page ｜ markets, trends, customers, returns ｜ charts 60–70%, commentary 30–40% ｜ 2–4 charts + 2–5 commentary blocks ｜ direct labels, orange event frames ｜ not suited to pure narrative.
+Financial model / comps table page ｜ history and forecast, comps, sensitivity ｜ main table 60–85%, commentary right or below ｜ 1 main table + 1 commentary area ｜ multi-level headers, light-yellow total rows ｜ not suited to abstract processes.
+Risk / plan matrix page ｜ risks–mitigants, status–actions ｜ left category column 15–20%, evidence in the middle, actions/timing on the right ｜ 4–8 rows ｜ colored side labels, probability or scoring color scales ｜ not suited to unprioritized long checklists.
+Proposition list page (occasional) ｜ multiple premises advancing in sequence ｜ deep-blue vertical column at left, numbered list stacked at right ｜ 4–6 items ｜ orange numbered circles ｜ not suited to dense data tables.
+
+【Density Baseline】
+Most body content occupies about 70–85% of the canvas. Argument pages: 3–6 evidence blocks. Chart pages: 2–4 charts, with columns, points, terminal values, and key differences labeled directly, forecast segments in light tints/hatching; explanation text in 2–5 blocks of about 40–120 words/characters equivalent each. Financial model pages center on one main table occupying 60–85% of the width, with 4–8 short comments to the right/below. Whitespace concentrates between title and subtitle, between modules, and in the outer margins; the body keeps no unexplained swaths of blank space — only section pages may hold large whitespace.

skills/open-kimi-ppt/reference/design_system/finance/lake-blue-memo/design.md

@@ -0,0 +1,75 @@
+# Lake Blue Memo · Finance STYLE DESIGN SYSTEM
+
+## PART A — Category Baseline (Universal Minimum)
+
+Institutional-grade investment research presentations benchmarked against top-tier fund research reports, due diligence materials, and strategy outlooks (global investment bank and sovereign wealth fund standards). Chart-heavy and text-dense, these decks should feel more like formal reports than consulting presentations: Excel-style financial model tables (three financial statements, scenario analyses, and sensitivity grids) carry the argument. Every figure must be traceable, every assumption explicit, and every conclusion actionable.
+
+Content and layout:
+1. Thesis-led structure: transaction decks follow “Executive Summary → Investment Highlights → Target/Market → Business Plan/Financial Diagnosis → Valuation and Returns → Risks and Mitigations → Decision Request”; research and outlook decks follow “Core View → Evidence → Scenario Analysis → Allocation Recommendation.” The ending must resolve into a clear decision or position.
+2. Tables take center stage: valuation ranges, trading comparables, three-statement financials, scenario analyses, and sensitivity grids should follow the row-and-column discipline of an Excel financial model, with clearly separated assumption columns, driver rows, and check rows. Charts serve the numbers—football-field valuation charts, IRR bridges, attribution waterfalls, and scenario fan charts—and are never decorative.
+3. High but orderly density: use multi-column layouts with a clear information hierarchy. Each page answers one investment committee question. Density comes from evidence modules, not walls of text.
+
+Visual style:
+1. Aesthetic: restrained, authoritative, and compliance-clean. Do not use lifestyle photography, decorative icons, or marketing superlatives.
+2. Typography: when institutional gravitas is required, use serif type for titles; use sans serif for table body text, chart labels, and all numbers. Always align numbers by column.
+3. Color: use a conservative institutional palette—navy/charcoal structural colors, with one restrained accent (such as gold or signal blue) reserved for key figures and recommended actions. Use one consistent red/green pair for positive and negative values, and only to convey financial meaning.
+4. Graphics: use fine table rules, precise vector chart lines, and thin bridge-chart connectors. Do not use 3D, shadows, or gradient fills as standard chart language.
+
+Structure and discipline (distilled from work we recognize as benchmark quality):
+1. Every content-page title must be a complete investment assertion (for example, “The asset will reprice because capacity is scarce,” not “Market Overview”), followed by a one-line subtitle defining scope and how to read the page.
+2. Every number must include a unit, point in time, and basis of measurement; estimates must be labeled (illustrative/internal estimate). Bare numbers are not allowed.
+3. Present returns only as scenario ranges (downside/base/upside, with probabilities or driver assumptions), never as a single-point IRR; every valuation conclusion must include a sensitivity table.
+4. Assumptions must be made explicit in a box on the page whose conclusion depends on them. Hidden assumptions are treated as defects.
+5. Pair every risk with its mitigation (Risk → Probability/Impact → Mitigation → Owner/Status). A page that presents risks without mitigations is defective.
+6. Every valuation page must have an anchor: trading comparables, precedent transactions, or DCF cross-validation. A multiple without a benchmark set is not evidence.
+7. Place parallel evidence in 2–3 equal-width columns; each numbered argument includes a one-line bold assertion plus one concise proof statement.
+8. Chart annotation discipline: label data marks directly, annotate bridge-chart blocks one by one, show forecast periods with dashed lines, and name key drivers inside the chart. Charts must never rely on legends alone.
+9. End with a decision or position: specify the requested action (approve/reject/approve with conditions) or a clear allocation stance (overweight/underweight/range), plus an action list and timeline. Do not end with a thank-you page.
+
+Default baseline (distilled from the layout skill; PART B may override): numbers first—key metrics (amounts/growth rates/multiples) must appear independently at a larger size than body text, immediately adjacent to their units and basis; retain at least one class of high-density financial table (three statements/comparables/sensitivity), with right-aligned numbers, thousands separators, and a consistent negative-number convention throughout; lock positive and negative semantic colors, using one fixed color for gains/positive values and one for declines/negative values throughout; forecasts and historicals must be visually distinct (dashed lines/light tints/hatched bands) and their basis labeled; place the data source, basis, and date range in small gray text at the bottom of every page, and never place unsourced figures in the primary visual area; maintain one visual character (dark navy/charcoal + metallic accents, or traditional white + blue-gray), selecting only one per deck.
+
+**Title Read-Through Test**: Reading only the page titles should tell the complete story; if they do not flow as a coherent narrative, the structure has failed.
+
+Density anchor: the specific minimum for each style—evidence modules per page, chart count, and annotation conventions—is defined by the 【Density Baseline】 reverse-engineered from that style’s source deck. This baseline does not impose a single universal number.
+
+### Global Prohibitions (AI Defaults · Mandatory)
+
+1. **No cards by default**: Unless the user explicitly requests them, do not use rounded rectangles or rectangular cards to create hierarchy or alignment. Containers with a genuine information function (status lights, assumption boxes, quotation boxes) are not considered cards. Lines, whitespace, type size, and weight differences are better tools.
+2. **No equal-split compositions**: Unless no other viable layout exists, do not default to three equal columns, four equal columns, or a 2×2 matrix. Equal division must be justified by the content structure (three scenarios naturally warrant three columns; frameworks explicitly named in the content are exempt).
+3. **No ungrounded AI color schemes**: The institutional palette specified in the body is exempt. Prohibited choices are generic, familiar, and unjustified defaults—blue-purple gradients, cyan-purple neon, rainbow flares, glass cards, and glowing borders.
+4. **No elements that clash with the overall style**: Do not introduce out-of-style treatments, such as rounded icons or rounded rectangles in a sharp-edged visual system.
+5. **Keep chart backgrounds transparent**: Use transparency or the page background color, except when reversing out on a dark background.
+6. **No empty regions**: Do not leave large blank areas in the body content zone. Every planned region must be filled with real content; an unfilled region is the most serious form of visual roughness.
+
+## PART B — Signature System (Template-Specific Signature; PART B Prevails in Case of Conflict with PART A)
+
+One-line style signature: a portrait institutional editorial-report aesthetic, grounded in paper white and deep lake blue, with grape purple as the title axis and bright blue-gray charts as evidence—dense without boxes, like a research journal designed to be browsed.
+
+【Color Palette】
+Background #FFFFFF occupies the vast majority of content-page area; near-white blue #EEF3F9 is reserved for full-page coverage on the opening statement page. Primary text #072535 is used for body text, running headers, and primary axis labels, keeping dense small type legible. Title color #490E5A is used for section names, pull quotes, and evidence conclusions, never as a body background. Primary accent #490E5A is reserved for judgment statements and evidence numbers; related deep purple #49004B is used only for small chart titles. Secondary #0C989A is used for large section numbers; #FFA500 marks the starting dot of a question-and-answer sequence; #DFE3F6 provides a light fill for grouped table headers. Chart series use #0229B1 for the primary series and #869198 for comparisons; #00A1C7, #5CA712, #490E5A, and #FFA500 join only for multi-series or phase encoding. Positive color #5CA712 and negative color #ED1C24 express directional judgments only. Excluding neutrals, standard pages use no more than 2 hues, and chart pages usually no more than 4; phase diagrams or heatmaps may exceed this for one-to-one data encoding, but the iron rule is that accent colors never flood a content page.
+
+【Layout Skeleton】
+Source-deck facts: the canvas is 612×792 pt, portrait Letter, with a 17:22 ratio; left and right body margins are approximately 54 pt. A lightweight running header and page number sit at the upper right, while a gray hairline spans the content width approximately 78 pt from the top. The body commonly uses a 48%｜4% gutter｜48% two-column grid, read vertically from the upper left to the bottom before moving to the right column. Chart-dense pages shift to a full-width primary evidence module at the top with paired left-right modules below, or two columns of small multiples. Cover only: full-bleed photography with a very large, white, left-aligned title. Contents page only: an index occupying approximately the left 2/3 and a deep lake-blue information sidebar occupying approximately the right 1/3. For 16:9 migration: on a 960×540 canvas, preserve approximately 5.5% outer margins, the top hairline, and the upper-right page number; reflow portrait long-form text into a 15%–22% title band above 2–3 horizontal evidence zones. Never scale the entire portrait page down proportionally and center it. Section pages retain giant, light-weight purple titles that may occupy 55%–62% of the left side, with the summary or first evidence module placed to the right and below.
+
+【Typography】
+Source-deck typography facts: the PDF embeds Ghost Regular, Light, Bold, Italic, and Medium, all sans serif (render with {latin: Liter, ea: MiSans}). Body text is 10 pt=1x; charts, tables, and sources are 8 pt≈0.8x; module headings and pull quotes are 18–20 pt≈1.8–2x; large content-page titles are approximately 38 pt≈3.8x; section titles and large numbers are 58 pt≈5.8x; the cover title is 85 pt≈8.5x. Use Light for giant titles and pull quotes, Regular for body text, Bold for evidence labels, and Italic only for terminology or subtle emphasis. For 16:9 migration: on 960×540, use 16–18 px body text, 13–15 px captions, 11–13 px sources, 28–34 px pull quotes, 56–64 px content-page titles, and 84–96 px section titles and large numbers. Preserve the original ratios and the relationship of “light titles, bold labels.” English titles use sentence case; section numbering is all caps. Keep numbers, units, and percent signs together without decorative tracking.
+
+【Chart Language】
+Useful chart types include line charts, grouped/stacked bar charts, stacked area charts, waterfall charts, scatter plots, phase-path diagrams, forecast/scenario tables, and correlation heatmaps; heatmaps and phase-path diagrams are both occasional-use formats. ⭐Begin every evidence module with “number + one-sentence purple conclusion,” followed by a hairline and a bold chart title; the chart itself receives no separate card background. ⭐A single page often arranges 2–4 small charts in two columns, each with its own short title, basis, and source, creating editorial-style small multiples. ⭐Use bright blue for the primary series and gray for comparisons; call out key inflection points with red outlined circles or thin arrows, and use only very light gray vertical bands for interval backgrounds. ⭐Forecast/scenario tables use light lavender grouped headers; only comparison cells receive pale cyan or pale warm-red fills, while white table bodies and fine lake-blue rules remain dominant.
+
+【Signature Components】
+Borderless purple pull quote: extract one judgment statement from the long-form text and place it at the bottom or side of a column, no more than 1 time per page, never overlapping a chart. Large cyan number + short gray rule above: use only for sections presenting a sequence of numbered viewpoints, with 1 per viewpoint and no more than 2 on a page. Green upward triangle/red downward triangle: use only for grouped positive and negative judgment lists, never as a replacement for ordinary bullets. Solid orange dot + question number: mark only the beginning of one question-and-answer sequence; do not repeat it on continuation pages. Photography is a hard constraint: the cover uses one full-bleed, shallow-depth-of-field natural macro photograph, with the title positioned over a low-texture area. Content pages must not substitute photography for data evidence; small portraits are limited to the contents-page information sidebar.
+
+【Prohibited】
+Do not use serif titles; in this template, the Ghost sans-serif family overrides PART A’s option for formal serif typography. Do not use large dark backgrounds with reversed body text on content pages; a solid deep lake-blue block is allowed only in the contents-page sidebar. Do not expand purple or blue into large content containers; they appear only in text, chart lines, table headers, or small signals. Do not center large content-page titles; section names, viewpoints, and questions are all left-aligned. Do not replace cross-supported small-multiple evidence with a single unannotated “hero image”; if one chart spans the full width, it must still include an evidence conclusion, basis, and source.
+
+【Slide Types and Layouts】
+Section opener｜Begin a new argumentative section｜Giant title in the top 30%, two columns or “text + table/chart” in the bottom 70%｜2–4 modules｜Section number, pull quote｜Not suitable for continuous data updates.  
+Two-column long-form page｜Explain one complex judgment progressively｜48% / 4% gutter / 48%, with space at the bottom of a column for a pull quote｜2–5 modules｜Purple pull quote｜Not suitable for comparing more than four entities at once.  
+Small-multiple evidence page｜Use multiple metrics to support one conclusion｜Full-width primary chart at the top or 2–4 charts in two columns｜3–6 modules｜Evidence-number strip, inflection-point circle/arrow｜Not suitable when there is only one qualitative viewpoint.  
+Numbered-viewpoint page｜Present 3–5 core judgments in sequence｜Approximately 50% charts and 50% numbering + explanation, or one judgment on each side｜4–6 modules｜Large cyan numbers, pull quote｜Not suitable for non-sequential content.  
+Positive/negative judgment list｜Present grouped trade-offs｜Two-column vertical lists, each item beginning with a triangle｜3–5 modules｜Red/green triangles｜Not suitable for neutral description.  
+Serialized Q&A page｜Develop an evidence-based answer around one question｜Question header approximately 15%, two columns below; continuation pages may include 1–2 charts｜3–6 modules｜Orange question dot, pull quote｜Not suitable for packing multiple unrelated questions onto one page.
+
+【Density Baseline】
+Source-deck content pages typically contain approximately 1,600–4,500 English characters. Long-form pages primarily use 2 columns, 4–8 natural paragraphs, and 1 pull quote; small-multiple pages contain 2–4 charts/tables and 1–2 explanation blocks, with an independent source for every evidence module. Bar charts usually label every bar; line charts label only key inflection points, extremes, and the latest point, rather than displaying numbers at every point. For new work in 16:9, recommended density is: 2–3 evidence blocks on section pages; 2–4 explanation blocks on long-form pages, each approximately 80–180 words/characters equivalent; 2–3 charts on small-multiple pages, increasing to 4 only when structurally necessary. Concentrate whitespace around giant titles, in the two-column gutter, between chart modules, and in the source area at the bottom of the page; do not leave large unassigned voids in the body content zone.

skills/open-kimi-ppt/reference/design_system/finance/prospect-annual/design.md

@@ -0,0 +1,68 @@
+# Prospect Annual · Finance STYLE DESIGN SYSTEM
+
+**Full creative commitment**: apply your complete creativity and imagination, treating every page as an independent, complete design work worthy of scrutiny. Structure, composition, charts, imagery, and visual narrative must be content-driven — never fall back on default layouts.
+
+**No AI slop**: no low-judgment, templated, decoration-stacked AI default output, including unjustified card arrays, evenly split grids, gratuitous blue-purple gradients or glassmorphism, icon walls, hollow oversized type, repeated skeletons, and visual gimmicks unrelated to the content. Every page must show clear design judgment with a basis in the content.
+
+**Landscape delivery**: the final presentation is uniformly 16:9 landscape. Portrait or non-16:9 source material serves only as style observation and must be recomposed into 16:9 landscape — the original canvas ratio must not be carried over directly.
+
+**Layout coordination**: style.md governs the final style rules (fonts / colors / components / chart language / prohibitions); the "layout / visual task" line that query.md gives for each page governs that page's content structure, primary anchor, and reading order — both are binding execution specs. Layout intent follows the query task line first; final visuals follow style.md.
+### Global Prohibitions
+
+1. **No default cards**: unless the user explicitly requests it, never use rounded rectangles or rectangular cards to build hierarchy or alignment — containers that serve an information function (status lights, assumption boxes, quote frames) do not count as cards. Rules, whitespace, and differences in type size and weight are the better tools.
+2. **No evenly split compositions**: unless no other layout is truly available, do not default to three-way splits, four-way splits, or 2×2 matrices — even splits must be justified by content structure (three scenarios naturally take three columns; frameworks explicitly named by the content are exempt).
+3. **No unjustified AI color schemes**: institutional palettes specified in the content are exempt; what is banned are bland, common, unjustified default colors — blue-purple gradients, cyan-purple neon, rainbow flares, glass cards, glowing borders.
+4. **No elements that clash with the overall style**: nothing outside the style system may appear — for example, rounded icons or rounded rectangles in a sharp style.
+5. **Transparent chart backgrounds**: transparent or the page background color, except reverse-white treatments on dark grounds.
+6. **Avoid large blank areas**: keep reasonable density in the body area; padding out or diluting content into sparse pages is treated as a defect. Whitespace for breathing, grouping, and rhythm is allowed — form follows content; never stuff material in just to fill the page.
+
+## PART A — Category Baseline (universal floor)
+
+Institutional-grade investment research presentations, benchmarked against top-tier fund research reports, due diligence, and strategy outlooks (bulge-bracket bank and sovereign wealth fund standards). Chart-heavy, text-dense, more report-like than consulting decks: arguments are carried by Excel-style financial model tables (three-statement models, scenario analyses, sensitivity grids). Every number traceable, every assumption explicit, every conclusion actionable.
+
+Content and layout:
+1. Thesis-driven structure: deal decks run "Summary → Investment highlights → Target/Market → Business plan/Financial diagnostics → Valuation and returns → Risks and mitigants → Decision request"; research outlooks run "Core stance → Evidence → Scenario analysis → Allocation recommendation". The ending must land on an explicit decision or stance.
+2. Tables are the protagonist: valuation ranges, comparable companies, three financial statements, scenario analyses, sensitivity grids, with Excel financial-model row/column discipline (assumption columns, driver rows, and check rows clearly distinguished); charts serve the numbers (football-field valuation charts, IRR bridges, waterfall attributions, scenario fan charts) and are never decorative.
+3. High but ordered density: multi-column layouts with clear information hierarchy; each page answers one investment-committee question; density comes from evidence modules, not walls of text.
+
+Visual style:
+1. Aesthetic: restrained, authoritative, compliance-clean. No lifestyle photography, no decorative icons, no marketing superlatives.
+2. Typography: use serif for titles when institutional gravitas is needed; sans serif for table body text, chart labels, and all numerals; numbers are always column-aligned.
+3. Color: a conservative institutional palette — navy/charcoal structural colors, one restrained accent (e.g., gold or signal blue) reserved for key figures and recommended actions; positive/negative signs use one consistent red-green pair, expressing financial meaning only.
+4. Graphics: thin table rules, precise vector chart lines, fine connector lines on bridge charts; no 3D, no shadows, and gradient fills are not a routine chart language.
+
+Structure and discipline (distilled from the artifacts we consider benchmarks):
+1. Every body-page title must be a complete investment assertion (e.g., "This asset will reprice on capacity scarcity" — not "Market overview"), with a one-line subtitle stating scope and how to read the page.
+2. Every number carries unit, as-of date, and basis; estimates must be labeled (illustrative / internal estimate); no bare numbers.
+3. Returns are given only as scenario ranges (bear/base/bull, with probabilities or driving assumptions) — never a single-point IRR; any valuation conclusion must be paired with a sensitivity table.
+4. Assumptions must be made explicit in boxes, placed on the page where the conclusion depends on them; hidden assumptions are treated as defects.
+5. Risks and mitigants appear in pairs (risk → probability/impact → mitigant → owner/status); a page with risks but no mitigants is a defect.
+6. Valuation pages must be anchored: comparable companies, precedent transactions, or a DCF cross-check — a multiple without a reference set is not evidence.
+7. Parallel evidence runs in 2–3 equal-width columns; numbered arguments each carry a one-line bolded claim plus one short proof sentence.
+8. Chart annotation discipline: data labels attached to marks, bridge-chart connector blocks labeled block by block, forecast segments dashed, key drivers named inside the chart; charts may not rely on legends alone.
+9. The ending must land on a decision or stance: items submitted for decision (approve / reject / conditionally approve) or an explicit allocation stance (overweight / underweight / range), with a to-do list and timeline — never close with a thank-you page.
+
+Default baseline (distilled from the layout skill; PART B may override): numbers first — key metrics (amounts / growth rates / multiples) must be set larger than body text and presented standalone, immediately adjacent to unit and basis; keep at least one class of high-density financial table (three statements / comps / sensitivity), with right-aligned numbers, thousands separators, and negative-number conventions unified deck-wide; positive/negative semantic colors fixed — up/positive and down/negative each use one fixed color that never changes across the deck; forecast and history must be visually distinguishable (dashes / light tints / hatched bands) and labeled with the basis; every page carries source, basis, and time range in small gray type at the page bottom — numbers without sources are not allowed in the main visual area; a single temperament per deck (dark navy/charcoal with metallic accents OR traditional white ground with blue-gray) — pick one per deck.
+
+Density anchor: each style's specific floor (evidence modules per page, chart count, annotation habits) follows the 【Density Baseline】 reverse-engineered from that style's source material — this baseline imposes no uniform numbers.
+
+
+## PART B — Signature System (this template's signature; where it conflicts with PART A, this section governs)
+
+One-line style signature: deep navy and generous whitespace as the ground, neon mint green and cool lavender purple setting the focus; on four-column research pages, giant condensed assertions and single-block colored conclusion panels create annual-report rhythm.
+
+【Color Palette】 Body background #FFFFFF; primary text and body titles #000533; full-bleed #000533 only on covers/section pages. Primary accent #1EFF8F for section display type, arrow tiles, and key columns; supporting #A29AFF for section display type, line charts, and decorative squares, diluted to #C7C2FF and #DAD7FF for quote panels or secondary columns. Conclusion panels use #A5FFD2; whole-block focus panels use #D2FFE9. The table of contents may exceptionally go full-bleed #78FFBC; occasional special features may use a #FBFFCC ground with #FFE900. Chart series prefer #001AFF, #1EFF8F, #A29AFF/#DAD7FF, adding #FFE380 and #998C00 as needed when stacking. Positive/negative colors inherit from PART A — the source has formed no stable increment. Each page is usually deep navy plus 1–2 hue families; multi-series charts are the exception; saturated accents must not blanket the body.
+
+【Layout Skeleton】 Source facts: canvas 1224×792, 17:11 (1.545:1); narrow margins on all four sides; four columns unfold at an even rhythm with narrow, consistent gutters. Title fixed upper-left; section openers hold an oversized assertion across roughly the top quarter with four columns below, and continuation pages go straight to four columns or "left two columns of text, right two columns of charts". Navigation appears only in the table of contents — numbered items with horizontal rules and three columns of information; the footer centers the running title and right-aligns the page number. Reading path: assertion → left-side explanation → right-side evidence charts → colored conclusion panel. Cover/section only: deep-navy ground, oversized title at left, and at right one arrow tile above a rounded-square mosaic. 16:9 migration: unify finally to 16:9 landscape, keep the four columns, the left-right division of labor, the footer, and the section mosaics; compress title-to-body spacing and pull the body forward; reduce line counts within the four columns; tall three-column panels become horizontal panels spanning 2–3 right-side columns.
+
+【Typography】 Source facts: the whole deck is sans serif; the PDF font table verifies display headings in `SharpGroteskBold-14`, body and charts in `Inter-Regular`, emphasis in `Inter-Bold`. Render: titles in Anton (ultra-condensed bold feel, via customFonts) or MiSans Bold; body/charts in Inter (via customFonts) or Liter (Chinese: MiSans). Body text stays compact and continuously readable; subheads and chart captions are only slightly stronger than body; main assertions and pull quotes form clear focal points; section slogans and cover slogans become full-page protagonists only on their special pages. No standalone KPI tier — column-top and endpoint numerals sit close to the chart-label level; footnotes recede noticeably. Section eyebrows are all caps, everything else uses sentence-style titles; no serif. In 16:9 landscape, keep the "ultra-condensed bold display titles + Inter body" hierarchy and weight relations — do not carry over the source's absolute sizes or ratios.
+
+【Chart Language】 Actually used: line charts, multi-series/dual-axis lines, clustered columns, column + dot dual-axis, stacked columns, and zero-crossing columns; no data tables in the body — when tables are required, inherit PART A. ⭐ Half-page evidence chart: chart title above, the chart occupying the right two columns, the left two columns holding explanation and conclusion panel. ⭐ Layered annotation: column charts label every column, line charts mark only endpoints/key values, legends identify series only. ⭐ Forecast bands are used only occasionally and are not mandatory to migrate. Backgrounds follow the page; sources sit directly under charts.
+
+【Signature Components】 Body pages stably use 1 — at most 2 — light rounded conclusion/quote panels in #A5FFD2, #C7C2FF, or #DAD7FF, spanning one to two columns, placed close to the text or chart they explain, never covering charts and never split into card walls. The cover and 4 section pages use the "arrow tile + rounded-square mosaic", hugging the right half / right edge only, never entering the body. The running title and page number are the footer anchors. Portraits appear only occasionally in the foreword.
+
+【Prohibited】 No serif titles — gravitas is carried by the ultra-condensed bold. No full-bleed deep navy/mint/purple on ordinary body pages — dark grounds go only to covers/sections, and bright-green full-bleed only to the table of contents. No replicating conclusion panels into card walls. No extending photos into body backgrounds. No per-point markers and numerals on line charts. No reusing the arrow mosaic outside section pages.
+
+【Slide Types and Layouts】 Big-title overview page ｜ section stance and first round of evidence ｜ top 25–30% giant assertion, four columns below ｜ 5–6 blocks ｜ conclusion panels, optional single right-side chart ｜ not suited to tiled multi-metric layouts. Two-column argument + half-page chart page ｜ text and one main chart cross-validating ｜ left 48% explanation, right 52% charts ｜ 4–5 blocks ｜ 1 conclusion panel + direct labels ｜ not suited to many small charts. Four-column continuation / multi-theme survey (occasional) ｜ parallel factors ｜ four narrow columns, 1–2 conclusion panels at bottom/side ｜ 5–7 blocks ｜ light-purple judgment boxes ｜ not suited to single-evidence magnification. Focused analysis panel page (occasional) ｜ one judgment with multiple explanation passages and one chart ｜ left column limits the text, right three columns hold a light-mint panel containing title, explanation, and chart ｜ 4–6 blocks ｜ combined charts inside panels ｜ not suited to unrelated topics. Full-bleed feature analysis (occasional) ｜ a single feature's history–drivers–outlook ｜ light full-page ground, main chart at left, explanation in the middle, conclusion at right ｜ 5 blocks ｜ single-color conclusion block ｜ not to be applied repeatedly.
+
+【Density Baseline】 The 12 body pages typically hold 4–6 independent blocks, with a few four-column pages reaching 7; the 8 chart body pages each carry exactly 1 chart. Explanation blocks run about 120–220 words/characters equivalent, main assertions 24–50, conclusion panels 30–70; column charts label every column, line charts mark only key points. Whitespace concentrates between title and body, in gutters, and above the footer; the content area keeps no orphan blocks. Covers/section pages are the low-density exception.

skills/open-kimi-ppt/reference/design_system/finance/rice-paper-annual/design.md

@@ -0,0 +1,62 @@
+# Rice Paper Annual · Finance STYLE DESIGN SYSTEM
+
+## PART A — Category Baseline (universal floor)
+
+Institutional-grade investment research presentations, benchmarked against top-tier fund research reports, due diligence, and strategy outlooks (bulge-bracket bank and sovereign wealth fund standards). Chart-heavy, text-dense, more report-like than consulting decks: arguments are carried by Excel-style financial model tables (three-statement models, scenario analyses, sensitivity grids). Every number traceable, every assumption explicit, every conclusion actionable.
+
+Content and layout:
+1. Thesis-driven structure: deal decks run "Summary → Investment highlights → Target/Market → Business plan/Financial diagnostics → Valuation and returns → Risks and mitigants → Decision request"; research outlooks run "Core stance → Evidence → Scenario analysis → Allocation recommendation". The ending must land on an explicit decision or stance.
+2. Tables are the protagonist: valuation ranges, comparable companies, three financial statements, scenario analyses, sensitivity grids, with Excel financial-model row/column discipline (assumption columns, driver rows, and check rows clearly distinguished); charts serve the numbers (football-field valuation charts, IRR bridges, waterfall attributions, scenario fan charts) and are never decorative.
+3. High but ordered density: multi-column layouts with clear information hierarchy; each page answers one investment-committee question; density comes from evidence modules, not walls of text.
+
+Visual style:
+1. Aesthetic: restrained, authoritative, compliance-clean. No lifestyle photography, no decorative icons, no marketing superlatives.
+2. Typography: use serif for titles when institutional gravitas is needed; sans serif for table body text, chart labels, and all numerals; numbers are always column-aligned.
+3. Color: a conservative institutional palette — navy/charcoal structural colors, one restrained accent (e.g., gold or signal blue) reserved for key figures and recommended actions; positive/negative signs use one consistent red-green pair, expressing financial meaning only.
+4. Graphics: thin table rules, precise vector chart lines, fine connector lines on bridge charts; no 3D, no shadows, and gradient fills are not a routine chart language.
+
+Structure and discipline (distilled from the artifacts we consider benchmarks):
+1. Every body-page title must be a complete investment assertion (e.g., "This asset will reprice on capacity scarcity" — not "Market overview"), with a one-line subtitle stating scope and how to read the page.
+2. Every number carries unit, as-of date, and basis; estimates must be labeled (illustrative / internal estimate); no bare numbers.
+3. Returns are given only as scenario ranges (bear/base/bull, with probabilities or driving assumptions) — never a single-point IRR; any valuation conclusion must be paired with a sensitivity table.
+4. Assumptions must be made explicit in boxes, placed on the page where the conclusion depends on them; hidden assumptions are treated as defects.
+5. Risks and mitigants appear in pairs (risk → probability/impact → mitigant → owner/status); a page with risks but no mitigants is a defect.
+6. Valuation pages must be anchored: comparable companies, precedent transactions, or a DCF cross-check — a multiple without a reference set is not evidence.
+7. Parallel evidence runs in 2–3 equal-width columns; numbered arguments each carry a one-line bolded claim plus one short proof sentence.
+8. Chart annotation discipline: data labels attached to marks, bridge-chart connector blocks labeled block by block, forecast segments dashed, key drivers named inside the chart; charts may not rely on legends alone.
+9. The ending must land on a decision or stance: items submitted for decision (approve / reject / conditionally approve) or an explicit allocation stance (overweight / underweight / range), with a to-do list and timeline — never close with a thank-you page.
+
+Default baseline (distilled from the layout skill; PART B may override): numbers first — key metrics (amounts / growth rates / multiples) must be set larger than body text and presented standalone, immediately adjacent to unit and basis; keep at least one class of high-density financial table (three statements / comps / sensitivity), with right-aligned numbers, thousands separators, and negative-number conventions unified deck-wide; positive/negative semantic colors fixed — up/positive and down/negative each use one fixed color that never changes across the deck; forecast and history must be visually distinguishable (dashes / light tints / hatched bands) and labeled with the basis; every page carries source, basis, and time range in small gray type at the page bottom — numbers without sources are not allowed in the main visual area; a single temperament per deck (dark navy/charcoal with metallic accents OR traditional white ground with blue-gray) — pick one per deck.
+
+**Title read-through test**: reading only the page titles across the deck should tell the complete story; if the titles do not read through, the structure has not been established.
+
+Density anchor: each style's specific floor (evidence modules per page, chart count, annotation habits) follows the 【Density Baseline】 reverse-engineered from that style's source material — this baseline imposes no uniform numbers.
+
+### Global Prohibitions (AI default routines · mandatory)
+
+1. **No default cards**: unless the user explicitly requests it, never use rounded rectangles or rectangular cards to build hierarchy or alignment — containers that serve an information function (status lights, assumption boxes, quote frames) do not count as cards. Rules, whitespace, and differences in type size and weight are the better tools.
+2. **No evenly split compositions**: unless no other layout is truly available, do not default to three-way splits, four-way splits, or 2×2 matrices — even splits must be justified by content structure (three scenarios naturally take three columns; frameworks explicitly named by the content are exempt).
+3. **No unjustified AI color schemes**: institutional palettes specified in the content are exempt; what is banned are bland, common, unjustified default colors — blue-purple gradients, cyan-purple neon, rainbow flares, glass cards, glowing borders.
+4. **No elements that clash with the overall style**: nothing outside the style system may appear — for example, rounded icons or rounded rectangles in a sharp style.
+5. **Transparent chart backgrounds**: transparent or the page background color, except reverse-white treatments on dark grounds.
+6. **No regional blank space**: the body content area keeps no swaths of blank space; any planned region must be filled with real content — failing to fill it is the most severe form of sloppiness.
+
+## PART B — Signature System (this template's signature; where it conflicts with PART A, this section governs)
+
+One-line style signature: a report-serious chassis overlaid with magazine-style highlight notes — warm paper white, ultra-bold black type, per-section color rotation, evidence dense but orderly.
+
+【Color Palette】 Background #F2EFE6; primary text/titles #000000; section accents rotate through orange #F9961F, pink peach #FFC1A8, bright green #08BC51, purple #D772FF, indigo #4C36FD — used only for highlights, key numbers, and a few series; section background colors #05225A/#4C1431/#05201B/#A4C9FF, special pages only; series colors #0B1636/#1058E2/#A4C9FF/#F6A031/#08BC51/#7421F7, fixed as solid fills; positive/negative inherit PART A. Body pages: 1 primary accent + ≤3 series per page; accents never blanket the page.
+
+【Layout Skeleton】 Source facts: 960×540, 16:9; roughly 40 pages carry a "color square + short label" at upper-left; titles are 1–2 line left-aligned assertions, with keywords set off by trailing rectangular highlights. Default: left 30–35% conclusion, right 65–70% main chart; complex pages use dual charts, chart + table, or three columns. Chart titles often carry a thin black horizontal rule; source/basis/time are pressed to the bottom; reading path: upper-left → main chart → footer. The cover and 9 section pages are solid dark or light blue, with a giant title at lower-left and a stepped stack of blocks + oversized numeral at upper-right — special pages only.
+
+【Typography】 Sans serif, family unverifiable; wide-set heavy letterforms, regular body, heavy titles, a little italic. Render: titles and giant numerals in Archivo Black (via customFonts) or MiSans Bold; body in {latin: Liter, ea: MiSans}. Body 1x, subheads 1.2–1.4x, assertions 2.2–2.8x, giant numerals 7–10x, footnotes 0.6–0.7x. Sections often in all caps; numbers carry units/thousands separators; highlighted phrases may share the same frame.
+
+【Chart Language】 Types: line charts, columns/stacked columns, stacked areas, horizontal bars, dual axes, bubbles, donuts, matrix tables, timelines. ⭐ Stacked charts use fixed solid colors with small legend swatches; ⭐ tables have dark full-row headers, with key rows fully saturated end to end; ⭐ dual axes pair light-tint columns/areas with a thin line; ⭐ common-baseline bubble comparisons (occasional) write values inside the bubbles; timelines run a horizontal axis with nodes alternating above and below (occasional).
+
+【Signature Components】 Title highlight blocks: 1–3 per page, may span lines without covering text. Section labels: once, at upper-left. Conclusion bubbles / number boxes: inside or beside charts, ≤2 per page, holding a single conclusion/number only. Stepped stacked blocks + giant numerals for section pages only; orange arrow capsules docked to chart titles, occasional.
+
+【Prohibited】 Body pages: no full dark/high-saturation grounds, no centered big titles, no sinking section stacked blocks into body pages; no gradients/glows/thick shadows; rounded corners only for conclusion boxes or functional entries; use photos, complex illustrations, and large screenshots cautiously.
+
+【Slide Types and Layouts】 Conclusion + main chart ｜ single primary evidence ｜ left 30–35% / right 65–70% ｜ 1 chart + 1–3 annotations ｜ highlights/number boxes ｜ not suited to multi-evidence. Full-width assertion + dual charts/table ｜ cross-validating comparisons ｜ two columns below ｜ 2–4 blocks ｜ highlights/solid table headers ｜ not suited to a single conclusion. Three-column evidence ｜ parallel facts ｜ three columns ｜ 3–9 short blocks ｜ labels/highlights ｜ not suited to long narrative. Timeline / process ｜ sequential relations ｜ horizontal axis with alternating nodes ｜ 8–12 nodes ｜ colored dots and dashes ｜ occasional use.
+
+【Density Baseline】 Default: 1 assertion + 1 main chart + 1–3 annotation boxes; explanations in 1–3 blocks of 20–60 words/characters equivalent each, with whitespace in the left column, above the title, and around charts. Dual-chart/table pages: 2 charts or 1 chart + 1 table, 2–4 blocks, labeling only key column points; three-column pages: 2–4 short facts per column. Timelines may be dense but the axis keeps whitespace; source/basis are fixed at the bottom — never fill space with decoration.