make-doc-skill

name

make-doc-skill

description

Generate company-specific business documents: board materials, IR / CFO proposals, internal approval documents, Chinese/Japanese bilingual documents, Word/PDF deliverables, and 1920x1080 HTML/PDF slide decks. Runs a Brain -> Router -> Canvas intelligence pipeline (not theme application): infers the real deliverable, rebuilds structure, challenges weak content, locks numbers and language, then renders and verifies real DOCX/PDF/HTML. Two modes: Quick Mode (drop a source file/text -> polished output) and Full Mode (8-step audited pipeline with company tokens, sparring, review, and PDF QA). Depth control S/M/L/XL. Triggered when the user wants to create, refine, or reformat a company document, board pack, IR/CFO proposal, business explanation, policy document, bilingual JP/ZH document, or a formal Word/PDF/slide deck — especially when a source company has its own logo, colors, language requirements, and compliance tone. For HTML visual one-pagers / diagram docs, an optional HTML output mode is available (see Reference Files). Requires document tooling for DOCX/PDF rendering; yt-dlp only if extracting from YouTube in HTML mode.

make-doc-skill v4 — Company Document Generator

Generate polished, company-specific business documents from reusable structure plus company brand tokens. The primary deliverables are formal Word/PDF board materials, IR / CFO proposals, bilingual Chinese/Japanese documents, and 16:9 HTML/PDF proposal decks.

This is not a theme applier. It runs a Brain -> Router -> Canvas intelligence pipeline that infers the real deliverable, rebuilds structure, fixes language and diagrams, applies per-company brand tokens, and then exports and verifies real DOCX/PDF/HTML.

Use when: the user wants to create, refine, or reformat a board material, internal approval document, IR / CFO proposal, business explanation, operating policy, bilingual JP/ZH document, or any formal Word/PDF/slide deliverable — especially when the source company has its own logo, colors, language requirements, compliance tone, and output folder conventions.

Reference Files

Methodology (loaded on demand):

• docs/PIPELINE.md — the 8-step audited Full-Mode pipeline (01_company_tokens -> 08_pdf_qa)
• docs/PROMPT_ENGINEERING.md — Brain/Router/Canvas role design + prompt meta-principles
• docs/DESIGN_RULES.md — DOCX / deck / diagram / token / brand-color hard constraints
• docs/SIZE_DEPTH_MATRIX.md — S/M/L/XL depth control

Optional HTML output mode (loaded only when the requested deliverable is an HTML visual one-pager or diagram doc, not a formal Word/PDF document):

• templates.md — 18 HTML visual template skeletons
• strategies.md — 28 content strategies + expert questioning chains
• components.md — 16 HTML component templates
• depth-rules.md — HTML depth-visualization rules

The HTML machinery is preserved as a capable secondary output. The company-document pipeline below is the primary path.

---

Core Methodology — Brain → Router → Canvas

The skill is a three-layer pipeline, not a single-shot generator. Even when Brain rewrites heavily, Canvas never dumps "analytical prose" into a template — Router forces the translation into document structure.

Layer	Role	Responsibility	Input	Output
Brain	Editor + domain analyst	Understand the real deliverable, rebuild structure, challenge weak content, fix language. This is the Codex Intelligence Layer.	Source material + company context	Structured, corrected content + section plan
Router	Document architect	Map deliverable type -> output family (formal DOCX / split bilingual / 16:9 deck / HTML) -> section schema + renderer + template	Brain output	Composition plan + renderer choice
Canvas	Renderer + typesetter	Apply tokens, layout rules, diagram rules; render DOCX/HTML; export PDF. Makes no editorial decisions.	Composition plan + tokens	DOCX/PDF/HTML + QA

Core principle: depth comes from the sharpness of judgment, not the number of steps. A board document is convincing because Brain read the material correctly and rebuilt it — not because Canvas applied a color theme. Tokens are the design memory; Brain judgment is the execution engine.

Prompt meta-principles (apply when writing any sub-prompt in this skill):

1. Role > Instruction — say "You are the CFO's board-document editor at a TSE-listed company," not "please improve this."
2. Constraint > Freedom — explicit prohibitions ("no fabricated figures") beat vague encouragement.
3. Structure > Prose — make the model fill a composition table, do not let it free-write the document.
4. Few-shot > Description — one correct DOCX/slide example anchors quality better than ten rules.
5. Segmented > One-shot — the pipeline exists so each step's prompt is short, focused, and auditable.

Two Modes — Quick Mode / Full Mode

Mode	Trigger	Flow
Quick Mode	User drops a source (existing Word/PDF/text) and wants a fast polished output, no project setup	Detect input -> extract -> infer real deliverable -> apply company tokens (or sensible default) -> Brain light pass -> render -> export PDF -> report path
Full Mode	Board/IR material, bilingual delivery, or "build the document properly"	8-step audited pipeline with intermediate files and user checkpoints (see docs/PIPELINE.md)

Quick Mode trades sparring depth for speed, but still runs numerical-fidelity and language-lock checks and never fabricates. After a Quick output, offer "upgrade to Full Mode" (keeps the normalized source, adds tokens/sparring/review).

Codex Intelligence Layer

Design tokens alone are not enough. They define colors, fonts, sizes, and spacing, but they do not decide whether a document is actually convincing, readable, or suitable for a Japanese listed-company context. This layer is the Brain in the pipeline above.

Use this layer every time:

• Infer the real deliverable, not only the literal file type. If the user asks for a Word/PDF, identify whether it is board material, IR material, CFO proposal, business explanation, or public-facing deck.
• Read the existing document before designing. Extract not only text, but also weak points: broken hierarchy, bad line breaks, crowded tables, unstable diagrams, wrong alignment, poor page rhythm, and mismatched tone.
• Convert vague brand direction into concrete tokens. For example, "do not use blue; use our corporate colors" means defining the brand's actual system (e.g. red/gray) and removing accidental blue from headings, tables, links, diagrams, and charts.
• Make editorial decisions. Rewrite headings, split overloaded paragraphs, normalize terminology, and tighten section order when the document needs it.
• Use document-native judgment. Formal DOCX needs chapter page breaks, readable 12pt body text, controlled tables, and stable TOC. A 16:9 CFO proposal needs visual pacing, slide hierarchy, and full-screen readability.
• Treat diagrams as communication tools, not decoration. Replace messy flow graphics with clean fixed SVG/PNG diagrams only when the diagram clarifies the business mechanism.
• Verify the real output surface. A generated DOCX is not done until the exported PDF is readable. An HTML deck is not done until browser rendering and PDF export both look correct.
• Preserve user file discipline. Do not overwrite source files. Put final files where the user expects them, especially the handoff folder (default ~/Downloads) or next to the original file when requested.
• Prefer decisive improvement over mechanical conversion. If the source was made by another model and is visibly weak, use the tokens as a base, then actively rebuild layout, hierarchy, language, and diagrams.
• Report only what matters: output paths, format status, unresolved assumptions, and any checks that could not be completed.

In short: tokens are the design memory; Codex-style judgment is the execution engine. The skill should inspect, decide, rebuild, verify, and deliver.

Sparring, Anchoring & Prohibition Rules

This is what makes the Codex Intelligence Layer reliable rather than just "an LLM rewriting." Run sparring at M depth and above; run anchoring and prohibition rules at all depths including Quick Mode.

Three-dimension expert questioning (pipeline Step 4)

Dimension	Goal	Questions to force
Hypothesis challenge	Test the core claim	Is the main claim defensible? Is the evidence primary data or assertion? What is the strongest counter-argument a board member or auditor would raise?
Blind-spot completion	Fill missing dimensions	What is missing — downside, risk, dilution, regulatory, tax, precedent, timeline? What will the audience ask that the draft does not answer?
Perspective reconstruction	Offer a better framing	Would risk-first / comparison-first / timeline-first framing serve this audience better than the current order?

Sparring rounds by depth: S = 0–1, M = 1–2, L = 2–3, XL = 3+ (XL adds an extreme test: "what if the core assumption is wrong?").

Anchoring rules (non-negotiable for listed-company material)

• Numerical fidelity — use figures exactly as in the source. Never round ¥2.4B to "about ¥2B," never invent a number. Currency symbols, units, and fiscal periods stay exact.
• Material anchoring — only state what the source supports. Mark missing data [DATA_GAP] rather than filling it.
• Source attribution — mark supplemental/background knowledge as [Supplemental] so it is distinguishable from sourced claims.
• Language lock — once the output language is set (Step 2), the whole chain obeys it. Quotations keep their original language; 03_source_normalized.md preserves source language; the DOCX language attribute and fonts follow the output language.
• Pyramid + MECE — conclusion first, then supporting arguments, then data. Sections must be mutually exclusive and collectively exhaustive.

Prohibited output (Brain pipeline)

• No vague filler ("this area is evolving rapidly," "we must monitor future developments") used to avoid stating a conclusion.
• No judgment avoidance — if the material supports a recommendation, state it.
• No fabricated figures, citations, legal/tax conclusions, or approvals. Unverified legal/accounting/tax statements must be explicitly marked draft/assumption.
• No phenomenon-listing without causal explanation.
• No surface-level paraphrasing of a weak source — rebuild it.

Required Inputs

Collect these before rendering:

• Company identity: company name, ticker, market, logo, corporate colors, official website, IR tone.
• Output type: DOCX/PDF, HTML/PDF deck, bilingual combined document, or split language files.
• Audience: board, CFO, IR, operating team, external partner, regulator, or public investor.
• Source content: Markdown, outline, existing Word/PDF, pasted notes, or bilingual draft.
• Language mode: Japanese, Chinese, bilingual side-by-side, or split Japanese/Chinese outputs.
• Delivery folder: default to ~/Downloads unless the user specifies another path.

If the source file is in Dropbox, Google Drive, Downloads, or a user handoff folder, read it as source material only unless the user explicitly asks to modify it.

Token Model

Every company should get a token file before rendering. Use JSON or YAML. Do not hardcode colors, fonts, sizes, or spacing directly into templates unless the token system explicitly allows it.

Minimal token shape:

company:
  legal_name: "Acme Holdings, Inc."   # example only — replace with the target company
  ticker: "0000"
  market: "Tokyo Stock Exchange Growth"
  brand_name: "Acme"

brand:
  logo_path: "assets/company-logo.png"
  primary: "#c0392b"        # example brand palette — replace with the company's real tokens
  primary_dark: "#922b21"
  primary_light: "#e74c3c"
  primary_bg: "#fdecea"
  text: "#242424"
  text_muted: "#666666"
  text_subtle: "#9a9a9a"
  background: "#ffffff"
  background_pale: "#fafafa"
  section: "#f3f3f3"
  section_alt: "#e9e9e9"
  border: "#d9d9d9"
  border_strong: "#bdbdbd"

language:
  ja_font: "Hiragino Sans"
  zh_font: "PingFang SC"
  latin_font: "Aptos"
  mono_font: "JetBrains Mono"

docx:
  page_size: "A4"
  margin_mm: 18
  body_pt: 12
  line_spacing: 1.18
  chapter_page_break: true
  toc_alignment: "left"
  page_number_alignment: "center"

slides:
  canvas_px: [1920, 1080]
  base_font_px: 20
  print_page_mm: [508, 285.75]

For a 1920x1080 HTML slide deck, use the same design-token discipline (example palette below — replace per company):

:root {
  --fs-display: 3.75rem;  /* 75px */
  --fs-title: 1.5rem;     /* 30px */
  --fs-heading: 1.1rem;   /* 22px */
  --fs-body: 0.9rem;      /* 18px */
  --fs-caption: 0.8rem;   /* 16px */

  --primary: #c0392b;
  --primary-dark: #922b21;
  --primary-light: #e74c3c;
  --primary-bg: #fdecea;
  --text: #272727;
  --text-muted: #737373;
  --text-subtle: #b2b2b2;
  --white: #ffffff;
  --bg-pale: #fafafa;
  --bg-section: #f3f3f3;
  --bg-section-alt: #ebebeb;
  --border: #dddddd;
  --border-strong: #c5c5c5;
}

Rules:

• Use only tokenized font sizes for slide text.
• Use company primary color as the only functional accent unless the company token set explicitly defines additional semantic colors.
• Avoid blue/green/orange functional colors when the company brand does not call for them.
• Use hierarchy through weight, spacing, and muted text, not uncontrolled font sizes.
• Keep color variables in :root; avoid hardcoded hex values in components.

Depth Control — S / M / L / XL

Depth controls how hard Brain analyzes and how rigorous QA is. It does not control whether numerical fidelity or language lock apply — those always apply. Full matrix in docs/SIZE_DEPTH_MATRIX.md.

Dimension	S	M	L	XL
Use case	1-page exec summary / memo	Standard proposal / explanation	Full board material / IR document	Comprehensive board pack + appendix
Section count	3–5	5–8	8–12	12+
Brain depth	Core conclusion + actions	+ evidence evaluation	+ explanatory framework + alternatives	+ multi-perspective review + extreme test
Sparring rounds	0–1	1–2	2–3	3+
Diagrams	0–1	1–2	2–4	4+
QA rigor	Output exists + fonts + path	+ TOC + pagination	+ bilingual parity + diagram QA	full Quality Gates + visual contact sheet
Deck slides (if 16:9)	3–6	6–10	10–15	15+

If the user does not specify a size, infer it from the deliverable: a memo is S, a CFO proposal is M–L, a full board material with appendix is L–XL.

Output Families

Formal DOCX/PDF (primary)

Use for board packs, internal proposals, policy explanations, and Japanese listed-company materials.

Rules:

• Body text must default to 12pt unless the user specifies otherwise.
• Japanese body font: Hiragino Sans or another approved Japanese corporate font.
• Chinese body font: PingFang SC or another approved Chinese corporate font.
• Latin/numeric font: Aptos, Arial, or the company-approved Latin font.
• Each chapter starts on a new page for formal board materials.
• Table of contents entries are left-aligned; page numbers should be visually consistent and centered or right-tabbed according to the chosen style.
• Tables should use restrained borders, company accent header fills, and readable 10.5-11.5pt table text depending on density.
• Do not use blue unless blue is part of the company brand tokens.
• Avoid Word freeform flow arrows for complex diagrams; render diagrams as fixed SVG/PNG images and place them into the DOCX.

Split Chinese/Japanese Documents

Use when the user asks to separate languages.

Rules:

• Generate one Japanese DOCX/PDF and one Chinese DOCX/PDF.
• Do not leave Chinese fragments in Japanese output or Japanese fragments in Chinese output unless they are proper nouns.
• Keep page structure, chart numbering, tables, and section hierarchy parallel across languages.
• Use language-specific fonts and proofread line breaks independently.

1920x1080 HTML/PDF Decks

Use for CFO proposals, IR decks, operating briefings, and large-screen review.

Base assumptions:

• Canvas: 1920x1080 px.
• html { font-size: 20px; }
• 1rem = 20px.
• Target: conference room readability at 1-3m and PDF export.
• Delivery: single HTML file plus PDF, zero build if possible.

Slide layout:

• Header accent bar: 6px using company primary color.
• Title band near top, content area below, footer fixed at bottom.
• Default content padding: 40px 88px.
• Compact content padding: 28px 72px.
• Tight content padding: 20px 64px.
• Footer height: about 48px.
• Use grids from 2 to 6 columns only; avoid 7+ column grids.
• Cards should have minimum useful width around 250px.
• Border radius should stay restrained, typically 6px or less.

Print/PDF:

@page {
  size: 508mm 285.75mm;
  margin: 0;
}

Rules:

• Do not use transform: scale() to fix print layout.
• Print with background graphics enabled.
• Verify PDF page count and page size after export.

HTML Visual Document (optional secondary mode)

For an HTML one-pager / diagram doc / visual explainer (not a formal company document), use the v3 HTML machinery in the reference files: templates.md (18 templates), strategies.md (28 strategies), components.md (16 components), depth-rules.md. That path produces a self-contained single-page-scroll or 16:9 HTML document. Apply company tokens instead of the default blue CI when a company is specified.

Document Schema

Use a content schema so company-specific rendering is repeatable:

document:
  id: "sample-board-material"
  title_ja: "新中期経営計画の概要に関する件"
  title_zh: "关于新中期经营计划概要的议案"
  document_type: "board_material"
  audience: "board"
  languages: ["ja", "zh"]
  outputs: ["docx", "pdf"]

sections:
  - id: "executive-summary"
    title_ja: "提案の趣旨"
    title_zh: "提案主旨"
    page_break_before: true
    blocks:
      - type: "paragraph"
        body_ja: "..."
        body_zh: "..."
      - type: "table"
        columns_ja: ["項目", "内容"]
        columns_zh: ["项目", "内容"]
        rows: []

diagrams:
  - id: "fund-flow"
    type: "flow"
    render_as: "svg_image_for_docx"

Per-section composition (Step 5 blueprint)

Before rendering, define each section with these fields (the Word/Deck analog of a visual composition table). This forces Router to translate Brain's analysis into document structure rather than prose:

1. Title — message-type, conveys the conclusion (not "Background" but "OA settlement is mainstream, but risk skews to the seller").
2. Core message — the one thing this section must land.
3. Block / component type — heading, paragraph, table, callout, KPI row, diagram.
4. Token / color allocation — which brand tokens; accent used sparingly.
5. Source reference — which normalized source material backs it.
6. Logical relation — how it connects to the previous/next section.
7. Layout hint — full-width table, two-column, diagram-left, etc.
8. Information density — low / medium / high (do not stack high-density sections consecutively).

Pipeline — Full Mode 8 Steps with Intermediate Products

Each step writes a reviewable file into the project folder and ends with a user checkpoint: proceed / redo this step / skip with defaults / jump to render. Skipping applies sensible company-token defaults. Iteration archives prior files to _history/v[N]/; the project root always holds the latest version. Full trilingual detail in docs/PIPELINE.md.

Step 1  Company tokens     -> 01_company_tokens.md     Brand/lang/docx/slide tokens (Router input)
Step 2  Document brief      -> 02_document_brief.md     Deliverable type, audience, language mode, depth (S/M/L/XL), outputs, delivery folder
Step 3  Source normalized   -> 03_source_normalized.md  Existing Word/PDF/text -> structured MD/YAML; original quotes preserved; [DATA_GAP] marked
Step 4  Codex sparring      -> 04_codex_sparring.md     3-dimension expert questioning, weak-point list, editorial decisions
Step 5  Composition plan    -> 05_composition_plan.md   Section-by-section blueprint (8 fields), renderer + template choice
Step 6  Layout review       -> 06_layout_review.md      Pre-render structure / quality / brand gate
Step 7  Render log          -> 07_render_log.md         DOCX/HTML generated, decisions logged, PDF exported
Step 8  PDF QA              -> 08_pdf_qa.md             pdfinfo / pdffonts + visual checks, gate results, final paths

The per-step actions reuse the Workflow below. Mapping: Workflow "Inspect / build tokens" = Steps 1–2, "Normalize content" = Step 3, "Choose renderer" = Step 5, "Render first output" = Step 7, "Run QA" = Step 8, "Place deliverables / report" = post-Step 8.

Pipeline transparency: every decision is recorded in its step file (traceable); the user can pause/review/modify at any step (auditable); Brain explains key choices in-line (explainable); iteration is cheap (low-cost rollback).

Workflow

1. Inspect source materials. Identify the current document, language, company, output expectations, and source folder. If an existing polished file exists, treat it as style evidence.
2. Build or update company tokens. Extract logo color, accent color, text colors, fonts, spacing, table rules, chart rules, and output placement rules. Keep company tokens separate from document content.
3. Normalize content. Convert existing Word/PDF/pasted text into a structured Markdown/YAML source. Preserve headings, tables, references, and legal/compliance wording.
4. Choose renderer. Use DOCX rendering for formal documents. Use HTML/CSS rendering for 16:9 decks. Use fixed SVG/PNG images for diagrams inside Word.
5. Render first output. Generate DOCX/HTML first, then export PDF from that source. Do not manually edit the PDF.
6. Run QA. Inspect page count, fonts, layout, TOC, line breaks, table widths, diagram placement, language residue, and whether brand colors are respected.
7. Place deliverables. Put final DOCX/PDF/HTML in the user-requested handoff folder. Default: ~/Downloads.
8. Report concise result. Give exact output paths and any remaining risks, such as unverified legal citations or missing official logo files.

Diagram Rules

Use diagrams only when they teach more than a paragraph or table.

Allowed diagram types: process flow, funds flow, architecture map, timeline, matrix, org tree, layer stack, waterfall, simple KPI chart.

Rules:

• One major diagram per slide or page section.
• For DOCX, create diagrams as SVG/PNG and insert as fixed images.
• For HTML decks, use inline SVG. Keep it self-contained.
• Avoid Mermaid unless the output pipeline reliably converts it into a fixed image.
• Use the primary color for at most 1-2 focal elements.
• Do not make every node red.
• Use neutral fills and borders for normal nodes.
• Keep node count under 9 where possible; split complex diagrams.
• Arrow endpoints must land exactly on node edges.
• Use consistent spacing multiples, preferably 4px/8px.
• Avoid decorative icons, glow, gradients, floating arrows, and non-brand colors.
• For funds flow, prefer a clean left-to-right or top-to-bottom flow with actor lanes, numbered steps, and a short note table.

Layout Rules For Formal Word Documents

Use these defaults unless the user asks otherwise:

• A4 portrait.
• Body: 12pt. Body line spacing: 1.15-1.2.
• Major heading: 17-20pt. Section heading: 14-16pt. Subsection heading: 12.5-13.5pt.
• Table text: 10.5-11.5pt. Caption/note: 9.5-10pt.
• Chapter starts on new page.
• Use clear hierarchy but avoid excessive decoration.
• Use company primary color for headings, rules, table headers, and small accent bars.
• Use neutral gray for borders and secondary text.
• Keep tables within page width; reduce text, rotate nothing unless unavoidable.
• Check that diagram-heavy pages do not have broken wrapping or overlapping objects.

Layout Rules For 16:9 Proposal Decks

Use these defaults:

• Size: 1920x1080. Each slide is a fixed viewport section.
• Titles are concise; do not overload title bands.
• Body blocks should fill the slide vertically; do not center sparse content as a substitute for adding useful structure.
• Use flex: 1, stretched tables, split columns, callouts, and notes to fill visual space properly.
• Avoid pure decorative backgrounds.
• Keep the company/product/topic visible in the first viewport.
• Make PDF print match browser rendering.

Language Rules

Japanese:

• Use board-level Japanese for listed-company documents.
• Prefer concise nouns in headings.
• Use Japanese punctuation and legal/compliance wording consistently.
• Avoid casual marketing language in board materials.

Chinese:

• Use professional Chinese for board/finance context.
• Keep company names, ticker codes, legal terms, and proper nouns stable.
• Avoid literal machine translation when Japanese governance terms require localized explanation.

Bilingual:

• If generating a combined bilingual document, align sections one by one.
• If generating split documents, render independently and QA each language.
• Language lock applies (see Anchoring rules): the chosen output language governs the whole render chain; quotations keep their original language.

Quality Gates

These are the concrete checks behind Step 6 (layout review) and Step 8 (PDF QA). Run them before final delivery:

• Output exists in the requested folder.
• DOCX opens without repair warnings.
• PDF exports from final DOCX/HTML source.
• Fonts are readable and language-appropriate.
• Body text is 12pt for formal Word documents unless overridden.
• TOC alignment is correct.
• Chapters start on new pages when required.
• Tables do not overflow page margins.
• Diagrams are not broken, cropped, or made from unstable Word shape arrows.
• No unintended blue/green/orange appears when the company brand system is red/gray.
• Chinese/Japanese split files contain the correct language.
• Page count is reasonable and matches TOC after updating fields.
• For HTML decks, browser screenshot and PDF export both show non-overflowing slides.
• For PDF decks, page size is 16:9 and all pages are present.
• Numerical fidelity holds: every figure, unit, currency, and fiscal period matches the source; gaps are marked [DATA_GAP]; unverified legal/tax statements are marked draft.

Useful local verification commands:

pdfinfo "/path/to/output.pdf"
pdffonts "/path/to/output.pdf"
python -m compileall scripts

For visual inspection, render PDF pages to images or contact sheets and inspect title pages, diagram pages, dense tables, and final appendix pages.

Evals

Acceptance cases the skill should pass are in evals/evals.json: branded Word refine, CN/JP split, CFO 16:9 deck, fund-flow diagram, numerical-fidelity/language-lock. Run them against real generated outputs, not just prompts.

Handoff Convention

Unless the user specifies another destination:

• Final files go to the user's Downloads folder — ~/Downloads (macOS/Linux) or %USERPROFILE%\Downloads (Windows) — unless the user specifies another destination.
• If refining a file from Downloads, also place the polished file near the original when requested.
• Keep output filenames close to the source filename plus a clear suffix such as _精修版, _日本語版, _中文版, or _CFO Proposal.
• Do not overwrite source files unless explicitly requested.

Questions To Ask Only If Needed

Ask the user when the answer cannot be safely inferred:

• Which company logo/color should be authoritative if multiple logos or colors exist?
• Should the document be public-facing IR style or internal board style?
• Should bilingual content be combined or split?
• Is the PDF expected to be A4 formal material or 16:9 deck material?
• Are legal, accounting, or tax statements already approved, or should they be marked as draft assumptions?

Do not pause for minor style choices when a reasonable company-token default is available.

Future Folder Structure

When converting this into a fully installable folder skill, use:

make-doc-skill/  (or company-document-generator/)
  SKILL.md
  references/
    brand-token-discipline.md
    prompt-modules.md         # sparring, anchoring, prohibition prompt blocks
  templates/
    docx-formal-board.yml
    html-1920x1080-proposal.html
    tokens.example.yml
  scripts/
    render_docx.py
    render_html_deck.py
    export_pdf.py
    qa_pdf.py
  evals/
    evals.json
  examples/
    company-board-docx/
    company-cfo-deck/
  html-mode/                  # the v3 HTML machinery, optional
    templates.md  strategies.md  components.md  depth-rules.md

Keep SKILL.md focused on workflow and rules. Put long templates, examples, and rendering code into separate files only when the skill is installed as a folder.